Search

The current measurement of soil properties can be limited in the variable-rate fertilizer application using soil analysis, due to the labor-intensive and time-consuming in the laboratory. Rapid, non-destructive soil detection is highly required to be developed using visible-infrared spectroscopy. Among them, Raman spectroscopy can be used to detect the vibrational properties of molecules, due to the fast, non-invasive, small sample treatment, and free interference from water in various fields. Nevertheless, incident light is also required to induce the Raman scattering, in order to compensate for the information with infrared spectroscopy. Therefore, the Raman scattering measurement can be widely used in the composition and dynamic process of soil minerals, bacteria colonies, and humic fraction in soil sensing. However, it is still lacking on the fluorescence effect of soil in the sensing of soil nutrients, which can weaken the Raman signals and the information extraction for the quantitative analysis. Raman spectroscopy has been introduced into some novel measurement techniques to improve the signal-to-noise ratio in recent years, including the shifted excitation Raman difference spectroscopy (SERDS), confocal Raman microscope, and surface-enhanced Raman spectroscopy (SERS). Meanwhile, both aliphatic and aromatic compounds are Raman active in the quantitative and qualitative detection of soil organic matter (SOM). Moreover, many phosphorous compounds in soil are also Raman active, leading to low prediction accuracy using visible-near infrared spectroscopy. In this review, the research progress was proposed on the Raman spectroscopy in the rapid detection of soil nutrients, together with the technical means in suppressing soil fluorescence interference to obtain high-resolution Raman signals. In the detection of SOM, the fusion of infrared and Raman spectral data significantly improved the prediction accuracy of 43% in the root mean square error (RMSE). In addition, the SERDS technique was used to detect the SOM of 33 soil samples. The reconstructed Raman peaks of soil minerals and organic materials obtained an excellent prediction accuracy of the SOM, with the determination of coefficient R 2=0.82, and the residual prediction deviation RPD=1.81. Raman spectroscopy was used to detect the water-soluble nitrogen in the soil solution, whereas, the SERS was to enhance the Raman peaks. An excellent correlation was obtained between the concentration of water-soluble nitrogen and the SERS data (R²=0.91). Research showed that Raman spectroscopy can be an effective tool to identify the different phosphate species in soil. Wavelet packet decomposition of Raman spectra was used to predict the phosphorus concentration in the phosphate-mixed soil with the SOM leached, where the accuracy of the regression model reached R²=0.94. Since the measurement area of Raman spectroscopy depended mainly on the spot size of the laser that irradiated on the sample surface, the spatial variability of the soil sample can be difficult to focus on the target substance. The effective Raman signal of soil nutrients can be obtained with high spatial resolution while suppressing the interferences from the background light. The combination of SERDS and micro-Raman technologies can be expected to serve as an in-situ measurement of soil nutrients. Reducing the redundant variables is crucial when applying spectral fusion since interpretability and reproducibility of the prediction model are paramount in the sensor development using Raman spectroscopy. [ABSTRACT FROM AUTHOR]

Smart farming for field crops is a significant part of the smart agriculture. It aims at crop production, integrating mod‐ ern sensing technology, new generation mobile communication technology, computer and network technology, Internet of Things(IoT), big data, cloud computing, blockchain and expert wisdom and knowledge. Deeply integrated application of bio‐ technology, engineering technology, information technology and management technology, it realizes accurate perception, quanti‐ tative decision-making, intelligent operation and intelligent service in the process of crop production, to significantly improve land output, resource utilization and labor productivity, comprehensively improves the quality, and promotes efficiency of agri‐ cultural products. In order to promote the sustainable development of the smart farming, through the analysis of the develop‐ ment process of smart agriculture, the overall objectives and key tasks of the development strategy were clarified, the key tech‐ nologies in smart farming were condensed. Analysis and breakthrough of smart farming key technologies were crucial to the in‐ dustrial development strategy. The main problems of the smart farming for field crops include: the lack of in-situ accurate mea‐ surement technology and special agricultural sensors, the large difference between crop model and actual production, the instan‐ taneity, reliability, universality, and stability of the information transmission technologies, and the combination of intelligent ag‐ ricultural equipment with agronomy. Based on the above analysis, five primary technologies and eighteen corresponding second‐ ary technologies of smart farming for field crops were proposed, including: sensing technologies of environmental and biologi‐ cal information in field, agricultural IoT technologies and mobile internet, cloud computing and cloud service technologies in agriculture, big data analysis and decision-making technology in agriculture, and intelligent agricultural machinery and agricul‐ tural robots in fireld production. According to the characteristics of China's cropping region, the corresponding smart farming development strategies were proposed: large-scale smart production development zone in the Northeast region and Inner Mon‐ golia region, smart urban agriculture and water-saving agriculture development zone in the region of Beijing, Tianjin, Hebei and Shandong, large-scale smart farming of cotton and smart dry farming green development comprehensive test zone in the North‐ west arid region, smart farming of rice comprehensive development test zone in the Southeast coast region, and characteristic smart farming development zone in the Southwest mountain region. Finally, the suggestions were given from the perspective of infrastructure, key technology, talent and policy. [ABSTRACT FROM AUTHOR]

Wheat basic seedling number is one of the most important sources of the total number of wheat ears. In turn, the leading factor can also dominate the wheat yield per unit area. It is an essential prerequisite for the timely and accurate acquisition of the within-field spatial difference information of the wheat basic seedling number. The variable-rate topdressing of nitrogen fertilizer can then be implemented in the manner of precision agriculture. The population density of wheat tillers can often be regulated to realize the fertilizer reduction with a better yield. Unmanned Aerial Vehicle (UAV) remote sensing imagery can be efficiently obtained at the field level in recent years. However, the vegetation and background features can be only processed without considering the influence of mixed pixels of the imagery in the traditional agricultural UAV remote sensing applications. The accuracy and reliability of wheat basic seedling number inversion cannot fully meet the large-scale production in smart agriculture. In this study, the quantitative inversion accuracy of wheat basic seedling numbers was improved using the mixed pixel decomposition model of UAV remote sensing imagery. Firstly, the UAV remote sensing imagery was acquired with a spatial resolution of about 2.5 cm using DJI Mini drone. The relative radiometric calibration was then completed using the invariant target method. Furthermore, the endmembers of vegetation and soil, as well as the mixed pixels were extracted from the reflectance image, which accounted for 2.23%, 0.28%, and 97.49% of the pixels, respectively. The spectral signatures were acquired for the endmembers of vegetation and soil using the reflectance values. Consequently, the decomposition model was established using mixed pixels of UAV remote-sensing images. The linear decomposition was used to divide each mixed pixel into 2 components of vegetation and soil. The abundance data was acquired for each component. The vegetation abundance model was used to calculate the Fractional Vegetation Coverage (FVC) of the experimental field. The proportions of vegetation endmember and abundance were then evaluated over the total area of “1m and 2 rows”. Finally, a linear regression model was established between the FVC and the ground truth data of 15 sets of wheat basic seedling numbers. The determination coefficient R² was 0.87. Besides, the regression model was verified using 3 other ground truth data of wheat basic seedling numbers. The verification results show that the Root Mean Square Error (RMSE) was 1.97 seedlings/m² . The higher inversing accuracy was achieved in this case, compared with the average wheat basic seedling number of 217.442 seedlings/m² for the wheat field. A comparative experiment was performed on the FVC thematic maps. The traditional vegetation index method was used, including the Visible-band Difference Vegetation Index (VDVI), Green Red Difference Index (GRDI), and Green Red Ratio Index (GRRI). The linear regression models were then established between each FVC of VDVI, GRDI, GRRI, and ground truth data of wheat basic seedling number. The determination coefficient R² and RMSE were calculated as 0.79, 0.56, 0.47, and 6.06, 7.04 and 4.43 seedlings/m2, respectively. Therefore, better performance was achieved in the quantitative inversion model of the wheat basic seedling number using the mixed pixels decomposition of UAV remote sensing images. The findings can provide data support for the precise variable topdressing of nitrogen fertilizer at the tillering stage of wheat. [ABSTRACT FROM AUTHOR]

Water is essential for plant growth, and water shortage will have an impact on plant yield, growth and quality, therefore, rapid and nondestructive detection of water content in maize leaves is of great significance for scientific guidance of irrigation, and it is also important for improving crop yield. In order to rapidly detect the moisture content of maize leaves, transmission spectrum on the band of 300-1 700 nm was selected to predict moisture content of maize leaves. The testing system included the light source and sensors. The light source part was provided by Wuling Optical Instrument Company. The transmission spectral range of 350-820 nm was obtained with ocean STS-VIS spectrometer, and the spectrum of 900-1 700 nm was measured by Wuling NIRez near-infrared spectrometer. Although the 820-900 nm transmission spectrum was absent, the transmittance curves obtained from different water gradients showed the same trend. Ten maize plants at stage V9 were detected. In order to eliminate the interference of different blade thickness on the experimental results, the first piece of fully expanded leaves from the top of the plant was selected as the experimental sample. The samples were taken back to the laboratory, and 10 cm long strips were cut along the veins from the middle of the leaves. The experiment measured and recorded transmission spectra curve and the moisture content of maize leaves under different water gradient. The moisture content measured by fresh weight moisture content formula, the transmission rate calculated according to Lambert-Beer law. The total number of experimental samples was 98, validation set was 33, and modeling set was 65. The total moisture content of maize leaves was 10%-85% and the average moisture content was 57.9%. In order to eliminate the influence of the noise caused by the spectrometer itself, the Savitzky-Golay method was used for pretreatment. The data was processed using correlation analysis between the transmission spectra and the moisture content, and principal co. As a result, the sensitive wavelengths at 800, 932 and 1 423 nm were selected by the correlation analysis, and the sensitive wavelengths at 478, 748, 1058 and 1323 nm were extracted by principal component analysis. To further improve the determination ability of the model, four sensitive wavelengths were selected, which were 800, 1 323, 1 058 and 1 423 nm. Using the combination of these four wavelengths, 12 vegetation index such as ratio vegetation index, difference vegetation index and normalized difference vegetation index were obtained. In the 12 vegetation indices, the modeling accuracy and prediction accuracy were significantly higher than other vegetation indices, so the DVI (1 423, 800) and DVI (1 423, 1 503) were respectively used to predict the moisture content. The results showed that DVI, which was derived from the combination of strong water absorption band (1 423 nm) and weak water absorption wavelength (800 and 1503 nm), could effectively restrain the disturbance caused by structural changes, and it could sensitively reflect the moisture content of maize leaves. Multivariate linear regression model was established based on DVI (1 423, 800), T1323 and T1058, the calibration R2 reached to 0.968 8, the validation R2 reached to 0.951 9, and the root mean square error of prediction was 0.061. The results showed that water prediction model established by transmission spectroscopy could provide efficient guidance for plant leaf water rapid detecting instrument development. [ABSTRACT FROM AUTHOR]

Accurate prediction of winter wheat chlorophyll content at seedling stage is important for guiding precision management in the field. In order to acquire chlorophyll content of winter wheat leaves, traditional detection methods require to squash winter wheat leaves and applying chemical methods, which would have bad influence on crops growth and cause unnecessary waste of time on some level. It is proved that the spectroscopy analysis is an effective method to predict chlorophyll content of winter wheat. However, the drift and offset of spectral baseline has a great influence on the predicting accuracy. So, this study was carried out to eliminate the influence of the drift and offset. The experimental farm was randomly divided into 70 different sampling areas in Xiaotangshan, Beijing, and the winter wheat leaves were collected on April 20th in the period of seedling stage. The visible and near infrared canopy spectral reflectance of winter wheat was measured by an ASD FieldSpec handheld spectroradiometer at seedling stage. The chlorophyll contents of sampling leaves were detected by the spectrophotometer in the laboratory on the same day. The obtained data of the canopy spectral reflectance and chlorophyll content were assembled for each region individually. The multiple scattering correction (MSC) was used on the bands of 325-1025 nm wavelength, because many scattering errors were introduced into the measured spectral data due to the physical factors. The MSC method first requires establishing an ideal spectrum of all samples, and modifying all the other samples of near infrared spectra on the basis of ideal spectrum to, and spectral reflectance changes with the content of chlorophyll components in the sample meet the direct linear relationship. The absolute intensity difference of spectral reflectance of winter wheat canopy was weakened after the MSC pretreatment, and then scattering effect was effectively reduced. Baseline shift and offset problems were resolved, and the correlation coefficients of spectral reflectance and chlorophyll content were increasing through the MSC pretreatment. Furthermore, genetic algorithm (GA) was proposed for sensitive band selection. GA is a high efficient and globally random search optimization method which simulates Darwin's evolution by natural selection and genetic mechanism of biological evolution. According to the principles of choosing high frequency bands as characteristic wavelength, 486, 599, 699 and 762 nm crop canopy reflectances were selected to calculate vegetation indices, including ratio vegetation index (RVI), difference vegetation index (DVI), normalized difference vegetation index (NDVI) and soil-adjusted vegetation index (SAVI). The correlation between each vegetation index and chlorophyll content of winter wheat was analyzed. It was found that the correlation between each vegetation index and chlorophyll content of winter wheat significantly increased after the MSC. The results showed that DVI and SAVI could refrain interference of soil background during seedling period, and the optimal parameters were DVI(762, 599), SAVI(762, 599), DVI(762, 699) and SAVI(762, 699), and the correlation coefficients were all above 0.6. The DVI(762, 699) and SAVI(762, 599) were selected to establish the multiple linear regression (MLR) prediction model and the least squares-support vector regression (LS-SVR) prediction model. The 70 winter wheat samples were divided into 2 groups, 50 samples for model calibration and the other 20 samples for model verification. The results of MLR showed the determination coefficient of the calibration model was 0.528 and that of the validation model was 0.487. In order to improve the precision of the forecast model, the LS-SVR prediction model was applied, and the determination coefficient of the calibration model was 0.681 and that of the validation model was 0.611. It showed that the fitting result was ideal. With the application of spectral technology, it provides a feasible method to detect the winter wheat growth status at seedling stage. [ABSTRACT FROM AUTHOR]

The counting of cotton rows is dynamically a premise of accurate positioning of a visual plant protection vehicle, which can ensure that the vehicle can traverse at the end of a field. In this study, cotton during plant protection periods was taken as the research object, which presented a dynamic counting method of cotton rows based on Histogram of Oriented Gradient (HOG) and Support Vector Machine (SVM). Hogs of cotton rows and inter-row background were extracted as feature descriptors of images, then they were classified by an SVM classifier. In this way, cotton rows could be identified through classifying cotton rows and inter-row backgrounds. 7 000 positive and negative training samples were respectively prepared for HOG characteristics of training samples. Among them, positive samples contained a complete outline of cotton lines in width, by contrast, negative samples contained soil among lines, plastic films and incomplete outline of cotton lines, etc. Besides, 1 000 positive and negative test samples were prepared to optimize parameters of the HOG-SVM model. By setting a sliding window to training samples, hog features of sample images were extracted, meanwhile positive and negative samples were marked for initial training on SVM classifier parameters. The initially trained SVM classifier was used to classify and test the samples, and those with wrong classifications were picked out for re-marking and re-training. Such retraining processes would continue until all test samples were correctly classified. Then the final HOG-SVM model was solidified for video detections. In recognizing processes, the model was used to slide the window in sequence images and to detect. In the sequence images, the HOG-SVM model built in the previous step was used by the sliding window. In this way, cotton rows in the video could be detected. The same rows of cotton could be detected for many times. To accurately identify cotton rows, Non-Maximum Suppression (NMS) was used to normalize target windows of detected cotton rows. All candidate target windows were scored with confidence and sorted from high to low, then the one with the highest score was retained. In two consecutive images, NCC (Normalized Cross-Correlation) values of target windows of front and back frames were solved to match the same rows of cotton, realized dynamic tracking of cotton rows, and further realized counting. Visual perspectives would have different effects on the upper and lower ends of images. There was an adhesion among cotton rows at the upper end of images, and those at the lower end of images were lack of seedlings and lodging. Such interferences had impacts on identifications of cotton rows. To solve this problem, the Region of Interest (ROI) was set for video sequence images in this study, and recognition processes of all cotton rows were within the ROI region with a very complex cotton field environment. This method had strong robustness to change in a camera's view angles and natural environment, such as to change in the light as well as shook of branches and leaves caused by winds, etc. In this algorithm, the recognition rate of cotton rows was higher than 90% under the condition in which winds were lower than level-2 and the average detection time per frame was lower than 32 ms, which met requirements of actual field operations. This dynamic counting method of cotton rows was used as a basis for determining transverse distances of visual plant protection vehicles at field ends, which could be used in other semi-structured rows to plant low and short crops. [ABSTRACT FROM AUTHOR]

[Objective]Select deep learning cherry tomato fruit recognition models to achieve rapid and accurate detection of greenhouse cherry tomato fruits. [Methods] The training data set was constructed by collecting the cherry tomato fruit samples and annotating the data. YOLOv4 and YOLO v4-Tiny algorithms were used for model training. The precision, recall, mean average precision and comprehensive evaluation index Fl value of the trained models were analyzed. [Results]The precision of cherry tomato fruit recognition prediction was 100% and 96.84% for the trained YOLOv4 and YOLOv4-Tiny models, the recall rates were 91.12% and 90.53%, average precision was 96.32% and 94.18%, and the comprehensive evaluation index Fl values were 0.95 and 0.94, respectively. [Conclusion] YOLOv4 algorithm model was obviously better than YOLOv4-Tiny, could achieve accurate cherry tomato fruit recognition. [ABSTRACT FROM AUTHOR]

The experiment was conducted to study the rapid detection method of soluble sugar content in head cabbage by near infrared spectroscopy, partial least squares (PLS) regression and stepwise regression (SR). The experiment collected a total of 71 samples of head cabbage. The spectral data were measured by the spectrometer of ASD FieldSpec® Handheld™ 2 made in ASD Company, USA, and the vitamin C (Vc) was measured by the 2, 6 two chloro indigo titration method. The kennard-stone (K-S) method was used to divide all samples into a calibration set and a validation set according to the given ratio. First derivative (FD) and second derivative (SD) pretreatment methods were used to improve the S/N ratio so as to find the better one from them. The coefficient of determination (R2), root mean squared error of cross validation (RMSECV), and root mean squared error of prediction (RMSEP) were used to evaluate models. The samples were divided into calibration set (60 samples) and validation set (11 samples) according to the ratio of 6:1 by K-S method. Six PLS models were established by using original, FD, and SD preprocessed spectral data of preflectance and absorbance, respectively. The modeling results showed that the spectral preprocessing method using FD could well improve modeling accuracy, the R2 of the validation set with which increased from 0.85 to 0.93, and this method was thought to be the better spectral data pretreatment method in this experiment. For the spectrum after the FD treatment, 5 confidence levels (0.95, 0.975, 0.99, 0.995, 0.999 5) were set up, and SR method was used to select modeling wavelength. With the increase of confidence level, the number of selected wavelength variables decreased from 14 to 3, and then 5 multiple linear regression models were established by 5 sets of preferred wavelengths corresponding to 5 confidence levels. The R2 values were 0.91, 0.83, 0.77, 0.7 and 0.61, respectively, and the RMSECV values were 2.362 2, 3.316 3, 3.841 1, 4.125 2 and 4.962 5 mg/100 g respectively for calibration set, while the R2 values were 0.82, 0.71, 0.72, 0.75 and 0.69, respectively, and the RMSEP values were 2.121 9, 2.983 6, 2.902 4, 2.751 6 and 3.021 5 mg/100 g respectively for validation set. According to the determination coefficient and the root mean square error analysis of the calibration set and the validation set, the multiple linear regression models corresponding to 8, 6 and 5 wavelength variables were established, and the model performance was consistent for the calibration set and the validation set. The average R2 was 0.78, the average RMSECV was 3.760 9 mg/100 g for calibration set, and the average R2 was 0.73, the average RMSEP was 2.879 2 mg/100 g for validation set, which meant that using fewer wavelength variables to predict Vc content in head cabbage was practical. So the visible / short wave near infrared spectra could be used for quantitative detection of vitamin C content in head cabbage. For the PLS model, it had many wavelength variables, the information was large, and so the detection precision was high. For the multiple linear regression models with 8, 6 or 5 selected wavelength variables, with the reduction of number of wavelength variables, the R2 decreased, which could yet provide technical support for the development of portable testing instrument. Using near infrared spectroscopy analysis technique, the prediction model for the content of Vc in cabbage was established, and it could improve the working efficiency of the internal quality evaluation of the cabbage. The research provides a new approach for quality evaluation of head cabbage. [ABSTRACT FROM AUTHOR]

In order to realize intelligent management of soil sampling in farmland, remote intelligent management system of soil sampling was designed based on the technologies including 3S (GIS, geographic information system; GPS, global positioning system; RS, remote sensing), ZigBee wireless communication, RFID (Radio frequency identification), the 4G (4th generation) communication system, and so on. The system consists of acquisition node, gateway of the coordinator, mobile terminal and remote management software. The acquisition node can obtain the geographic location information of soil samples, the RFID electronic tag and the temperature and humidity of the soil environment; key technologies used in acquisition node are ZigBee wireless communication protocol, serial port communication protocol, and modules of data parsing, packaging, sending, and so on. The gateway of the coordinator is constituted by ZigBee coordinator and 4G module, realizing data transformation from the ZigBee wireless network to 4G network. And the data of acquisition node would be transmitted to management software of remote server when the 4G DTU (data transfer unit) was configured with server IP (Internet Protocol), port number and other configuration information. PDA (personal digital assistant) is used as mobile terminal, which has been remodeled by embedding a micro ZigBee coordinator and a management software in the system. So it is convenient and portable to achieve monitoring and controlling of soil samples collection in time for user, and another important function of PDA is it can realize the transformation of the data from soil samples acquisition node to remote server management software through GSM/GPRS (Global System for Mobile communication / General Packet Radio Service) network if there is no 4G network in remote farmland area. The 4G network is primary communication means under normal circumstances because of its higher efficiency and reliability. Therefore, this method can enhance reliability of the whole system with the double communication links. Management software of remote server develops the data reception, Baidu map, automatic data mapping (two-dimensional, three-dimensional) and other function modules based on Web, SQL (structured query language) Server, Socket and other technologies. Data will be saved to different groups for each test in data reception module, so it is convenient to process data for user. The real-time tracking of soil sampling can be realized through Baidu map with latitude and longitude information from GPS module, and the spatial distribution figure of soil sampling can be automatically generated by calling data collected in database or uploading the local experiment data. We can use system to collect soil sample information and obtain according total nitrogen content of soil sample in laboratory at the same time. In this way, total nitrogen content data of soil sample were imported into information bar by corresponding RFID label in soil management software, generating spatial distribution figure of soil nitrogen, which will provide the decision support for subsequent variable fertilization. The stability test of the system shows it has a high reliability, the packet loss rate of which is 0.2%. So the remote intelligent management system for soil sampling can work stably and accurately, and can be used in production practice. With the further study of the system, it will continue to be improved. The next step will be to increase the number of ZigBee acquisition nodes, and sensor is integrated with different functions of field information detection. [ABSTRACT FROM AUTHOR]

Crop spectrum characteristic analysis plays an important role on identification of crops, production estimation, condition monitoring, and nutrition diagnosis crop management. In order to predict the nutrient content of crop non-destructively and quickly, a reflectance spectrum detection system for winter wheat was developed to measure the reflectance of 350-820 nm. The system has three parts: optical sensor, data storage and transmission module and the controller. The optical system was developed based on the Ocean Optics STS-VIS sensor. The sensor was controlled by an enhanced electronic device to obtain reflected light with a high-sensitivity 1 024 pixel linear CCD array detector. The spectral information collection and analysis software was developed on Windows 7 platform, using PHP. Software included three modules: acquisition parameters, acquisition control and data management. After the controller and sensor connected successfully, users should optimize the system parameters, such as the integral time, the scan average times and boxcar width. Then it needed to choose the collection function, DN value. Finally, the reflectance data could be analyzed and stored. In order to test the performance of the spectrum analyzer, calibration experiment was carried out. A gray calibration board with four different gray gradations was involved. The correlation was analyzed between the spectral reflectance measured by spectrum analyzer and ASD Field Spec Hand Held 2. The result showed that the average correlation coefficient value was 0.94. it also showed that the developed spectrum analyzer was worked with good stability under different light conditions. The system was used to collect 350-820 nm (visible-near infrared) reflectance of winter wheat at heading stage in the field to detect the chlorophyll of winter wheat. In order to decrease the noise influence, canopy reflectance spectra curves were pre-processed by 1-order differential method. And the data smoothing was conducted by Bior Nr.Nd biorthogonal wavelet packet analysis method. After that, Monte Carlo sampling method was used to remove 5 outliers. And 8 sensitive wavelengths (719、572、562、 605、795、527、705、514 nm) were chosen using Random frog algorithm. In order to indicate the effective of preprocessing method, the chlorophyll content detecting PLSR (partial least squares regression) model was established based on original reflectance spectra. The modeling determination coefficient was 0.70 and predictive determination coefficient was 0.10. Meanwhile, the revised chlorophyll content detecting model was established. The involved data was pre-processed after 1-order differential and wavelet packet decomposition. The modeling determination coefficient was 0.69. The root mean square error of model was 1.364 8. And predictive determination coefficient was 0.52. The root mean square error of prediction was 1.839 7. The modling results showed the background interference and noise of wheat canopy reflectance were removed effectively. The system based on miniature spectrometer could estimate the chlorophyll of wheat leaves and help to diagnose the crop nutrition of winter wheat at heading stage. [ABSTRACT FROM AUTHOR]

Compared with the traditional plant-protecting machine, plant protection unmanned aerial vehicle (UAV) has advantages of high efficiency, high performance, good precision and good spraying effect. Most importantly, the promotion of physical protection is very significant. In order to fully guarantee that the plant protecting assignments can be allocated scientifically and the plant protecting UAV resources can be deployed efficiently, as well as for meeting the needs of plant protecting spray jobs, a high efficient algorithm for matching UAVs with plant protecting jobs was designed, and a high effective plant protecting assignments scheduling system was developed. The system can not only provide the UAV users with appropriate matching plant protecting assignments, but also help to find the appropriate matching UAVs for the users who need to rent the specific UAVs to carry out plant protection spray. In this paper, the existing technologies were analyzed and compared to find out which one could be used to fulfill matching algorithm for UAVs and assignments, the features of plant protection assignment for UAV were clarified, and a matching algorithm for UAV and its plant protection assignment based on the R-tree spatial indexing was designed. The R-tree is a completely dynamic spatial index of data structure, and sub-algorithms such as node inserting, deleting and querying operations are mutually independent. The matching algorithm for plant protection assignment includes the algorithms of inserting plant protecting assignment into the R-tree, querying plant protection assignment from the R-tree, and deleting some assignment when it is finished or canceled. By using the matching algorithm, the plant protection assignment intelligent recommendation system was developed, and it mainly included the region querying function and intelligent recommendation function. The region search function allows UAV users to search all the plant protection assignments within the scope of any rectangle dragged on the map. Meanwhile the intelligent recommendation function based on the user's current location can recommend the UAV users with the plant protection assignments nearby which meet the UAV's spray features, and also can help the users who have the specific plant protecting assignment to find the right UAVs to rent. We built the entire system by using Django web framework and programmed with Python language and JavaScript language. The results of system test showed that the matching algorithm for UAVs and their plant protection assignments based on the R-tree could handle more than 2 000 concurrent requests at the same time even in a lower configuration server. The querying algorithm's response time was less than 1 ms when processing a single request. The test of plant protection job inserted into the R-tree in batches showed that inserting 1 000 jobs took less than 1 s, and thanking to the dynamic nature of the R-tree, we could read and write the R-tree at the same time, so the insertion of the R-tree did not affect the query operation. It illustrates that the algorithm is flexible, accurate, dynamic and highly efficient, and it can be used to match UAVs with the corresponding appropriate plant protecting assignments reasonably and intelligently. [ABSTRACT FROM AUTHOR]

Crop yield information is the precondition for the implementation of precision agriculture, and it is the key to organize farmland production and achieve decision-making for agricultural management. Aiming at collecting yield data by the grain yield measurement and control system in harvest field and transmitting the collected data to the yield server in real time, and meanwhile meeting the demands of data access to all kinds of terminals, a data collection and service platform for monitoring grain yield was developed. It is helpful to achieve data real-time interacting among the on-site yield monitoring system, the mobile terminals and the remote yield server. Socket and SOAP (simple object access protocol) technologies are used to implement data exchanging between the remote server and various heterogeneous platforms. A grain yield data collecting system was developed and embedded in the IPC (industrial personal computer) mounted on the combine harvester to collect yield data in real time, and its modules included device setting, data collecting, data parsing and data uploading. The yield data are composed of the information input by user (farm and plot identification, which are input once for each plot), the global position system (GPS) data (longitude, latitude, altitude and time), the data (harvester speed, harvester cutter width, impulse voltage, grain moisture and grain temperature) from the sensors mounted in the harvester and the CAN (controller area network) bus. The collected yield data are packaged every 1 second and transferred to the yield server through Socket via GPRS (general packet radio service), and then they are parsed according to the yield data exchanging protocol and stored into the yield database established in the server. Then the yield data can be calculated, processed and visualized according to various modules embedded in the yield data collection and service system which is developed using Hadoop and Web GIS (web geographic information system) technologies. And the modules include user management, news management, bulletin management, document management, farm management, plot management, yield data management, harvester tracing and Web GIS service, and the grain yield data collecting system is connected with the remote data access terminals through the data collection module and data service module which are embedded in the yield data management. Yield data service is implemented using Web Service based on SOAP for data sharing for various terminals (such as mobile, tablet PC (personal computer), PDA (personal digital assistant), and desktop) with different operating systems (such as Windows, Macintosh, Linux or Android). A grain yield querying system was established for testing the yield data service XML (extensive makeup language) interface provided by the yield data collection and service system, the information including farm, plot and yield could be accessed smoothly, and the yield data visualization could be achieved accordingly. For improving the yield data transmission efficiency, GZIP (GNUzip) is used to compress SOAP messages and reduce bandwidth consumption. In addition, the digital signatures and encryption technology are used to guarantee the security of data transmission via internet. The system experiments were carried out, and the results showed that the grain yield data collecting system worked well on-site and the yield data could be collected and uploaded every 1 minute; the data collection and service system could receive, calculate, process and visualize the real-time yield data efficiently; and the Web Service interface could push the yield data querying results quickly according to client requests. The study indicates that the yield data interaction and sharing between heterogeneous platforms can be achieved by using the yield data collection and service platform proposed in this paper. [ABSTRACT FROM AUTHOR]

A rapid and accurate detection of a maize plant has been widely determined the food safety and quality assurance in modern agriculture. However, there is a great challenge on the complex environment of canopy leaves at the jointing stage. It is also necessary to select a feasible network model suitable for the precise identification and positioning of maize plants in the field. In this study, a novel device was developed for canopy recognition during the maize jointing stage using deep learning. A total of 3000 images of maize in the seedling and jointing stages were also taken to train the model of deep learning. Furthermore, a labeling strategy was proposed for the sample set, where the whole maize object was replaced with the center of the maize plant. Some data enhancement techniques were utilized to increase the robustness and generalization ability of the identification model for maize plants, including the passive brightness, chroma, rotation, mirroring, and blurring. After that, three datasets of maize plants were achieved, including 10 500 training sets, 3 000 cross-validation sets, and 300 test sets. In the model training, the SSDLite-MobileDet network model was first trained on the Google Colab cloud platform, and then compared with the SSDLite-MobileNetV3 model. Finally, an optimal SSDLite-MobileDet model was achieved with a speed of 110 frames/s and an accuracy of 92.4%. More importantly, some strategies were proposed to improve the performance of the model. 1) The specific procedure of extraction was visualized for the maize images during model training using a Convolutional Neural Network (CNN). It was found that the target information was significantly enriched in the output feature mapping, as the number of CNN layers deepened gradually. 2) The heart of the plant was labelled to avoid the serious leaf crossing at the jointing stage of maize. Thus, the novel model presented an average recognition accuracy of 92.4% under severe leaf crossing and different lighting conditions. 3) A rapid detection device was built for the maize crops with Raspberry Pi 4B+Coral USB, including Raspberry Pi 4B control, accelerator fast calculation, camera image acquisition, power, and image display module. As such, a rapid and accurate platform was obtained to collect, process, and display the images of the maize canopy in the field. 4) The model that trained on the GPU was quantized to transplant the trained model to the device. The data type of FP32 was converted to INT8, thereby ensuring that the quantized model occupied less memory. 5) A software was designed to run the SSDLite-MobileDet lightweight model on Raspberry Pi 4B, further to realize the acquisition, recognition, and display of maize images. The specific operation included system initialization, model loading, images acquisition, model processing, maize canopy recognition, positioning, and display. Finally, a field experiment was carried out to evaluate the rapid detection device for the maize crops, where the frame rate of the detected video was more than 89 frame/s, and the application recognition accuracy rate reached 91%. The findings can also offer strong support to the high-precision diagnosis and refined operation of maize in the field. [ABSTRACT FROM AUTHOR]

An accurate modeling of three-dimensional (3D) point cloud has been a crucial step to extract the physical phenotyping parameters of plants, such as the plant height, leaf quantity, and leaf area. In this study, an automatic registration of point clouds with multiple calibration balls was proposed to realize the 3D modeling of plants under a complex background in the field. A low-cost depth sensor (such as Kinect V3) was also selected to capture the images. The performance of the registration was then evaluated using the multiple angles of several plants under the indoor and in-field scenes. Four procedures were included the point cloud filtering and down sampling, multiple calibration balls extraction, correspondence points matching, as well as the calculation and registration of transformation matrix. 1) The pass-through filtering with only several boundary thresholds was used to reduce the noises, while the bounding box compression was used to down sampling the point clouds. 2) A Random Sampling Consistency (RANSAC) was also used in the multiple calibration balls extraction. Furthermore, a concept of point cloud subtraction was proposed to combine with the RANSAC, in order to form an automatic extraction of multiple calibration balls. Among them, the RANSAC was utilized to extract one single object at one time. As such, the center coordinates of each calibration ball were then calculated as the featured points. 3) The distances of any two feature points were calculated in the correspondence points matching. An automatic matching of correspondence points using the distance information was also proposed to realize the self-matching of the calibration balls. 4) The singular value decomposition was adopted to solve the transformation (rotation and translation) matrix, and then the registration of two pieces of point clouds was realized eventually. The experiments were carried out in two scenes: one was the indoor scene with the flat and clean surface, and another was the field scene with the uneven and complex conditions. Five plants were chosen as the research objects, including one sugarcane (indoor), one sorghum (indoor), and three young banana plants (in-field). The Kinect V3 sensor was utilized to capture the point clouds of each object in five orientations (the horizontal orientation of 0º, 90º, 180º, 270º, and one vertical orientation). A commercial point cloud processing software named LiDAR360 was adopted for comparison with the man-aid registration. After that, the object point clouds (horizontal 90°, 180°, 270°, and vertical) of each plant was used to implement the registration into the coordinate system of the source point cloud (horizontal 0°). The registration performance was evaluated to calculate the axial errors and point errors of the transformed coordinates, as well as the source coordinates of the centers of all the calibration balls. The results showed that the registration of different point clouds was successfully implemented in the different orientations, no matter in the indoor or in-field environment. Particularly, the generated 3D model of the plant was clear in shape. Specifically, there were an average axial error of 5.8-17.4 mm and an average point position error of 13.1-28.9 mm for the registration in the different scenes, similar to the registration in the manual registration software LiDAR360. In the required calculation time, the automatic matching took about 50 s on average for the process from the extraction of calibration sphere to the registration, whereas, the LiDAR360 took about 150 s with the manual selection of the correspondence points, indicating the efficiency increased by 67%. Consequently, it can be widely expected to serve a high-precision and automatic registration of plant point clouds acquired by low-cost depth cameras in complex field environments. The finding can provide a low-cost feasible solution for the 3D modeling and the extraction of phenotyping parameters of field plants. [ABSTRACT FROM AUTHOR]

To monitor the potato growth status, the chlorophyll content was involved and its vertical distribution was detected at different leaf positions in plants. In this paper, the experiment was conducted during flowering stage of potato in the greenhouse of School of Water Conservancy and Engineering, China Agricultural University in June 2017. The potato leaf samples were randomly collected at 3 leaf positions (upper, middle, lower) of different potato plants. Then the hyper-spectral data of 66 potato leaf samples were divided into 400 regions of interesting (ROIs) and the SPAD (soil plant analysis development) values of the corresponding leaf positions were measured, of which 140 were taken from the upper part of the leaves and both 130 from the middle and lower leaves. The chlorophyll distribution was analyzed based on the vertical leaf position. The results indicated that its spectral reflectance in the range of 382-700 nm increased with the rise of the position. In the range of 700-1019 nm, the reflectance at the lower leaf position was higher than that at the upper and middle leaves, and the mean values of SPAD were 36.41, 43.11, and 47.04, respectively, at lower, upper and middle position. The black and white corrections of the potato leaf sample images collected using the hyperspectral imaging system were performed. After extracting and calculating the average leaf spectrum of the chlorophyll measurement area, the spectral data were pretreated by the standard normal variable (SNV) correction method, and then 2 sensitive wavelength selection methods were applied to built the chlorophyll content estimation models. The 12 wavelengths were chosen by using the correlation coefficient (CC) analysis and the 11 wavelengths were selected by the random frog (RF) algorithm. The results from the partial least squares regression (PLSR) model showed that 12 sensitive wavelengths selected by the CC analysis method were mainly located in the range of 530-550 and 706-708 nm. And in the PLSR, the modeling determination coefficient was 0.7588, and the predictive determination coefficient was 0.5773. Meanwhile, based on the RF algorithm, 11 sensitive wavelengths were 554.62, 560.26, 575.04, 576.35, 595.09, 604.7, 649.44, 731.8, 752.78, 786.38, and 789.97 nm. The modeling determination coefficient of the PLSR prediction model built with these wavelengths was 0.842 3, and the predictive determination coefficient was 0.767 6. Thus, the chlorophyll content of each leaf of potato was calculated by RF-PLS model, and the visual distribution of chlorophyll content in potato leaves was plotted. The results showed that the hyperspectral imaging could reflect the dynamic response of potato chlorophyll in flowering stage and achieve the non-destructive detection of potato leaf chlorophyll content and the visual expression of chlorophyll distribution. [ABSTRACT FROM AUTHOR]

During the detecting and locating process of green tomatoes based on obtained visible images, problems such as shadow caused by uneven illumination source, occlusion of stems and leaves, and occlusion between fruits, need to be solved. In this study, a machine vision algorithm was put forward, which aimed to determine fruit location and size of green tomatoes. Normalized cross-correlation function (NCC) is a feature detection method using template matching. The algorithm firstly detected the potential location of green tomatoes through fast normalized cross-correlation function (FNCC). Then a gray histogram based classifier was used to classify if the location corresponded to green fruit. The histogram based classifier was built based on fruit areas and non-fruit areas of a certain size (30×30 pixels) extracted from the obtained image. Seven texture features, including mean, standard deviation, smoothness, third moment, uniformity, entropy, and gray level range were calculated for those fruit areas and non-fruit areas. Three classifiers including k-nearest neighbor (KNN), SVM (support vector machine), and Naive Bayes, were used to classify fruit and non-fruit areas using those 7 texture features as input vectors. SVM was chosen based on its performance. The non-fruit location was filtered out, and the number of fruit locations was estimated. Meanwhile, the image was segmented based on color analysis. Red and Blue component from RGB (red, green, blue) image, and Hue component from HSV (hue, saturation, value) image transformed from RGB images, were used as the basis for color analysis. Using the fruit potential location number estimated using FNCC as an input parameter of circular Hough transform (CHT), CHT was then applied to the edge image of the segmented result. The center coordinates and radius value of each circle were calculated. Finally, the detection results were merged based on the analysis of the distances of 2 centers of fruit circles detected using CHT. If the distance between 2 circles is smaller than the minimum fruit radius, the circle with a larger radius will be kept, while the other circle will be flagged as repeatedly detected one. Thus, the recognition and positioning of the green tomato were realized. When green fruits and red fruits appear on the same image, a red fruit detection algorithm based on the local maximum value method and random circle round transform detection, which was proposed by the author in another paper, would be carried out on the obtained image. Then the red fruit detection result was combined with the green fruit detection result. The proposed method combined the texture, color, and shape information of the image, and eliminated the disturbance of the color similarity between the green tomatoes and green leaves and stems. A total of 70 images were used in this study, in which 35 images were used as training images, and the other 35 images were used as validation images. The correct detection ratio for 72 fruits in the training dataset was 89.2%, and the correct detection ratio for 105 fruits in the validation dataset was 86.7%. The proposed method has provided a reference for the development of tomato harvesting robots for both red and green mature tomato fruits. [ABSTRACT FROM AUTHOR]

Stem diameters of maize are important phenotype parameters and can characterize the crop growth and lodging resistance, drawing more attentions from breeders. Traditional measurement about stem diameters is usually manual measurement, which is timeconsuming, laborious, and subject to human error. In order to rapidly measure stem diameters of maize in field, a method based on RGB-D (red, green, blue - depth) camera was proposed in this paper to extract stem diameters of maize. The color images and depth images of the maize plants at the small bell stage were captured by a RGB-D camera in field. First, maize stem was extracted by processing the color image. It was hard to recognize maize just according to the color differences in red, green and blue component between maize and background due to the illumination variations. To solve the problem, the component that represented the difference between green signals and illumination brightness was calculated and applied to segment maize with Otsu algorithm, and the binary image of maize was generated. And then erosion operation was conducted within region of interest to cut off the connection between little leaves and maize stem, and small regions were eliminated to remove weed and little leaves. The largest region of maize was saved after dilation operation. After that, skeletonization was conducted for main stem. There were crossing points at the points of contact between leaves and stem, and ending points at the points of contact between ground and stem, and the potential measurement region of stem could be identified by searching crossing points and ending points. The color coordinates of the potential measurement region were saved and corresponding point cloud data were generated based on the mapping relationship between color coordinate, depth coordinate and camera coordinate. Second, stem diameters were calculated by processing point cloud data. Noise points affected measurement accuracy of stem diameters, and K-nearest method was applied to remove scattered points from point cloud data. Then the filtered point cloud data of potential measurement region were clustered. There were some point cloud data on the edge of stem due to the measurement of time of flight (ToF), which were background noises. K-means method was used to divide the filtered point cloud data into 2 groups, and only the group whose central point was nearer to the camera was saved to represent maize stem. The saved point cloud data were one side of stem, and ellipse fitting based on least square method was carried out for the point cloud data. Long axis parameter and short axis parameter of ellipse were calculated respectively to indicate the stem diameters of maize. 20 samples were tested to verify aforementioned method, and the experimental results showed that the method proposed in this paper had a good performance in segmenting and identifying maize stem, though ellipse fitting method needed to be improved. The mean errors, standard deviation and mean relative errors of measuring stem diameters were 3.31 mm, 3.01 mm, 10.27% for long axis and 3.33 mm, 2.39 mm, 12.71% for short axis, respectively, indicating that the proposed method could be applicable for plant phenotyping. [ABSTRACT FROM AUTHOR]

In order to improve the performance of the agricultural machinery automatic navigation system, an automatic navigation path searching method of agricultural machinery based on GNSS (global navigation satellite system) was proposed. According to different demands of farm working, the system could generate the straight line or curve path for agricultural machinery automatic navigation according to users setting. In order to get the straight navigation path, the user should drive the tractor and record the current position marked as point A, and then choose the position marked as Point B at least 10 m far away from Point A. The straight line presupposed navigation path could be obtained by connecting Point A and B and extending the segment AB. The way of obtaining the curve presupposed navigation path is similar to the straight path searching method; the curve fits with several segments, and every segment is analyzed with the straight path searching method. When the navigation task began, the system would compare the current position and heading information of the tractor with the presupposed path to get the lateral deviation and heading deviation. In addition, a pure pursuit mode based on preview points research was proposed for steering control. The method didn't involve the complicated control theory, so that it could adapt the navigation system better. In the aspect of turning control, arcuate turning and pyriform turning patterns were selected as the major research objects. The turning path could be generated by the navigation system according to the tractor working width and the minimum turning radius after the users chose the kind of turning pattern, and a series of points could be chosen according to the tractor speed and each point was evenly spaced. When the navigation task began, the searching radius and preview point should be set according to the speed of the tractor. There were several points on the default navigation path falling in the searching circle; the point with the largest ID (identification) number would be selected as the preview point, and then the path of the tractor arriving to the preview point and the control turning angle would be obtained. To verify the path search method and the model of pure tracking performance, a tractor automatic navigation software was designed and implemented. The industrial computer as the carrier of navigation software, processed the GNSS data, IMU (inertial measurement unit) data and PLC (programmable logic controller) data, and then generated the corresponding decisions. A John Deere tractor was used as the platform for experiments, and the straight line / curve navigation experiments based on GNSS positioning technology were designed. The results of experiments were as follows: In the straight line navigation experiments, when the speed of the tractor was 0.8, 1.0 and 1.2 m/s, the root-mean-square error was 3.79, 4.28 and 5.39 cm respectively; in the turning navigation experiments, when the speed of the tractor was 0.6 m/s, the root-mean-square error of the arcuate turning navigation was 25.23 cm and the root-mean-square error of the pyriform turning navigation was 14.42 cm; for the comparison experiment, the root-mean-square error using the proposed method and fuzzy control method was 4.30 and 5.95 cm respectively in straight line navigation module, and 13.73 and 21.40 cm respectively in curve navigation module. The path searching method and the pure pursuit mode based on the researching of preview points can satisfy the requirement of the farmland works effectively. [ABSTRACT FROM AUTHOR]