Search

Currently, the main methods for detecting plant diseases are sampling and manual visual inspection. However, these methods are time-consuming, laborious and prone to misinterpretation. Rapid advances in Deep Learning (DL) techniques offer new possibilities. This study focused on analysing the confounding factors among three types of wheat rust (stripe rust, leaf rust and stem rust) and aimed to achieve higher classification accuracy. The following approaches were used: (1) Images were collected from several crops and diseases: Wheat Rusts Dataset (WRD), Wheat Common Disease Dataset (WDD), and Common Poaceae Disease Dataset (PDD); (2) Seven common convolutional neural network (CNN) models were made and their performance compared. DenseNet121 was selected as the base model, and its classification results further analysed. The results of the above analyses were then considered using phenotypic morphology and model structure analysis, as well as potential confounder discussions; (3) Adjustments and optimisations were made based on the identified confounding factors. The final improved model, designated Imp-DenseNet, achieved the following accuracies with different datasets: Top-1 accuracy = 98.32% (WRD), Top-3 accuracy = 97.30% (WDD) and Top-5 accuracy = 96.60% (PDD) (Top-x Accuracy refers to the accuracy of the top-ranked category that matches or containing the actual results). The study revealed the potential factors contributing to the confusion among the three wheat rusts and successfully achieved higher accuracy. It can provide a new perspective for future research on other diseases of wheat or other crops. [Display omitted] • Confusion factors of wheat rust were analysed from different perspectives • Combine knowledges of plant protection, plant phenotype morphology analysis, DL and CV • Remodel DenseNet using attention mechanism, SPP, and factorisation thoughts • Verify validity of methods with Grad-CAM, Yolo v8 and Transformer [ABSTRACT FROM AUTHOR]

The acquisition of maize tassel phenotype information plays a vital role in studying maize growth and improving yield. Unfortunately, detecting maize tassels has proven challenging because of the complex field environment, including image resolution, varying sunlight conditions, plant varieties, and planting density. To address this situation, the present study uses unmanned aerial vehicle (UAV) remote sensing technology and a deep learning algorithm to facilitate maize tassel identification and counting. UAVs are used to collect maize tassel images in experimental fields, and RetinaNet serves as the basic model for detecting maize tassels. Small maize tassels are accurately identified by optimizing the feature pyramid structure in the model and introducing attention mechanisms. We also study how mapping differences in image resolution, brightness, plant variety, and planting density affect the RetinaNet model. The results show that the improved RetinaNet model is significantly better at detecting maize tassels than the original RetinaNet model. The average precision in this study is 0.9717, the precision is 0.9802, and the recall rate is 0.9036. Compared with the original model, the improved RetinaNet improves the average precision, precision, and recall rate by 1.84%, 1.57%, and 4.6%, respectively. Compared with mainstream target detection models such as Faster R-CNN, YOLOX, and SSD, the improved RetinaNet model more accurately detects smaller maize tassels. For equal-area images of differing resolution, maize tassel detection becomes progressively worse as the resolution decreases. We also analyze how detection depends on brightness in the various models. With increasing image brightness, the maize tassel detection worsens, especially for small maize tassels. This paper also analyzes the various models for detecting the tassels of five maize varieties. Zhengdan958 tassels prove the easiest to detect, with R2 = 0.9708, 0.9759, and 0.9545 on 5, 9, and 20 August 2021, respectively. Finally, we use the various models to detect maize tassels under different planting densities. At 29,985, 44,978, 67,466, and 89,955 plants/hm2, the mean absolute errors for detecting Zhengdan958 tassels are 0.18, 0.26, 0.48, and 0.63, respectively. Thus, the detection error increases gradually with increasing planting density. This study thus provides a new method for high-precision identification of maize tassels in farmland and is especially useful for detecting small maize tassels. This technology can be used for high-throughput investigations of maize phenotypic traits. [ABSTRACT FROM AUTHOR]

• A novel model (LabTNet) is proposed to estimate crop biochemical traits. • LabTNet jointly estimates multiple wheat traits by hyperspectral data. • LabTNet exhibits superior predictive performance than existing methods. • Grad-CAM analyzed hyperspectral attention regions of LabTNet. Accurate estimation of crop leaf and canopy biochemical traits, such as leaf dry matter content (Cm), leaf equivalent water thickness (Cw), leaf area index (LAI), dry leaf biomass (DLB), leaf total water content (LW), and fresh leaf biomass (FLB), is essential for monitoring crop growth accurately. The vegetation spectral feature technique combined with statistical regression methods is widely employed for remote sensing crop biochemical traits mapping. However, the crop canopy spectral reflectance is influenced by various crop biochemical traits and uncertainties in geometric changes of light and soil background effects. Consequently, the remote-sensing estimation of crop biochemical traits is limited. A potential solution involves training a deep learning model to understand the physical relationship between crop biochemical traits and canopy spectral reflectance based on a physical radiative transfer model (RTM). The primary focus of this study is to propose a winter-wheat leaf and canopy biochemical traits analysis and mapping method based on hyperspectral remote sensing, utilizing a deep learning network for leaf area index and leaf biochemical traits deep learning network (LabTNet). This study consists of four main tasks: (1) Field-based measurements of winter-wheat spectra and biochemical traits were conducted in two growing seasons. A PROSAIL RTM was also employed to generate a simulated dataset representing comprehensive and complex winter-wheat field conditions. (2) The LabTNet deep learning model was pre-trained using the simulated spectra dataset to acquire knowledge of the physical relationship between crop biochemical traits and canopy spectral reflectance derived from the RTM. Subsequently, the model was re-trained using the field-based spectra dataset from two growing seasons, employing a transfer learning technique. (3) An analysis was conducted to assess the performance of LabTNet against traditional statistical regression methods in estimating crop leaf and canopy biochemical traits. The study used the gradient-weighted class activation mapping (Grad-CAM) technique to analyze the attention regions of input spectra (454:8:950 nm, 960:10:1300 nm, 1450:10:1750 nm, 2000:10:2350 nm) by different convolutional neural network layers in LabTNet, aiming to enhance the interpretability of deep learning models. (4) Winter-wheat leaf and canopy biochemical traits (Cw, Cm, LAI, DLB, LW, and FLB) were mapped using the LabTNet deep learning model. Our research has the following conclusions: (1) Combining the RTM and deep learning techniques yields higher winter-wheat biochemical trait estimates than traditional statistical regression methods. (2) Different LabTNet deep learning model layers focus on distinct areas of canopy reflectance, corresponding to the sensitive regions for various winter-wheat biochemical traits. (3) LabTNet demonstrates similar winter-wheat leaf and canopy biochemical traits estimation performance using visible and near-infrared (VNIR) reflectance data and full-spectral (FS) range hyperspectral reflectance as inputs (Cw: R 2 = 0.603–0.653, RMSE = 0.0015–0.0015 cm; Cm: R 2 = 0.511–0.560, RMSE = 0.0006–0.0007 g/m2; LAI: R 2 = 0.773–0.793, RMSE = 0.65–0.66 m2/m2; LW: R 2 = 0.842–0.847, RMSE = 67.93–70.73 g/m2; DLB: R 2 = 0.747–0.762, RMSE = 21.10–21.89 g/m2; FLB: R 2 = 0.831–0.840, RMSE = 86.26–90.30 g/m2). The combined use of UAV hyperspectral remote sensing and the LabTNet model proves effective in providing high-performance winter-wheat leaf and canopy biochemical trait maps, offering valuable insights for agricultural management. [ABSTRACT FROM AUTHOR]

The accurate large-scale measurement of peach crowns is vital in horticultural science and the optimization of orchard management. Nowadays, numerous crown parameters (e.g., crown area, height, and volume) can be obtained via the analysis of point clouds or photographs. Current laser-based sensors provide the required reliable and accurate information; however, they are costly and time-consuming. Therefore, a simpler approach for crown measurement is required. For this purpose, this study presents a pipeline for the monitoring and clustering of 259 peach tree crowns based on unmanned aerial vehicle (UAV) images of a peach orchard in Southeast China. Considering the limitation that the original aerial image dataset contains little information, a data augmentation process is adopted, and an efficient deep learning architecture based on conditional generative adversarial networks (cGANs) was designed to extract the crown area. Then, the shape of the crown area was clustered using an edge detection process and a $k$ -means algorithm. Finally, an ellipsoid volume method (EVM) was applied to estimate the crown volume. Five indicators—namely, $Q_{\mathrm {seg}}$ , $S_{\mathrm {r}}$ , Precision, Recall, and F-measure—were employed to evaluate the crown extraction effects, and the average results for testing samples were 0.832, 0.847, 0.851, 0.828, and 0.846, respectively. Compared with other approaches—namely, fully convolutional network (FCN), U-Net, SegNet21, the excess green index (ExG), and the color index of vegetation extraction (CIVE)—the proposed cGAN model performs better, achieving an accuracy improvement of 5%–25%. For the estimation of crown volume, using measurements from a light detection and ranging (LIDAR) scanner as a reference, the correlation coefficient and relative-root-mean-square error (R-RMSE) were found to be 0.836% and 14.93%, respectively. Overall, the results demonstrate that the proposed method is feasible for measuring peach tree crowns. The wide application of such technology would facilitate applied research in plant phenotyping and precision horticulture. [ABSTRACT FROM AUTHOR]

Fast and accurate counting of wheat ears in field conditions is a key element for determining wheat yield. To obtain the number of wheat ears in a field, we propose a new counting algorithm based on computer vision. This algorithm counts wheat ears in remote images through semantic segmentation regression network (SSRNet). SSRNet is a multistage convolutional neural network that we propose to achieve counting problems through regression. In SSRNet, first, the original image is cropped to increase the amount of data. This method effectively solves the small sample dataset. Next, based on the cropping results, we build a fully convolutional neural network (FCNN) to segment wheat ears in field conditions. FCNN increases the accuracy of wheat ears counting by accurately segmenting wheat ears in a complex background. Then, we build a regression convolutional neural network (RCNN) to count wheat ears based on the segmentation results of FCNN. In RCNN, we propose a new activation function positive rectification linear unit (PrLU) to process the last layer of the fully connected layer, so that RCNN can effectively count the number of wheat ears in the image. Finally, a counting strategy is proposed to count the number of wheat ears in the original image. To verify the counting performance of SSRNet, we compare the counting result of SSRNet with the real value of manual statistics. The results show that the average accuracy (Acc), $R^{2}$ , and root mean squared error (RMSE) of the SSRNet count results on the test set in this article are 0.980, 0.996, and 9.437, respectively. It can be seen from the results that our proposed method can accurately count wheat ears in field conditions. At the same time, the counting time (0.11 s) shows that SSRNet can quickly estimate the number of wheat ears in field conditions. We concluded that this study can provide important technical support for the high-throughput field wheat ears counting task in large-scale phenotyping work. [ABSTRACT FROM AUTHOR]

• Introduction of a novel model that integrates OBIA and the attention mechanism for precise orchard mapping. • The contribution of SAR is studied in orchard classification. • The model obtains more accurate classification results than RNN-based and CNN-based deep learning models. • The uncertainty of the super-pixel size on classification accuracy is investigated. Reliable and accurate classification of orchards is important for the dynamic monitoring of large-scale orchards and food security evaluation. At present, the very similar spectral profiles of different fruit trees and the high susceptibility of optical data to interference by weather conditions limit the resolution of orchard classification. Synthetic Aperture Radar (SAR) imagery provides an advanced solution to this problem, with the advantages of being immune to weather conditions and advances in deep learning techniques. This paper presents an orchard classification model (STCM) with optical and SAR fusion, which integrates the advantages of the Simple Non-Iterative Clustering (SNIC) super-pixel algorithm with a deep learning algorithm based on a multi-headed attention mechanism, and it achieves the best classification accuracy of above 0.82 in comparison with existing deep learning models. Meanwhile, the model has exceptional classification accuracy and robustness in a study area with fragmented plots, many fruit tree species, and various distributions of optical images. In the absence of optical data, the classification accuracy of the STCM model using only SAR data is around 0.70, which makes the model have a promising potential application value. This study provides a technical solution for accurately obtaining different orchard categories in high-resolution remotely sensed orchard images. The study helps to improve the management and operation of orchards and provides a strong basis for decision-making in the fruit industry to enhance the sustainability of the global fruit industry. [ABSTRACT FROM AUTHOR]

Timely determination of weed distributions in fields is crucial for the precise spraying of herbicides. This facilitates weed control while saving costs and protecting the environment. Existing weed detection strategies often rely on the utilization of numerous weed samples to train detection models directly, which presents challenges in situations involving limited weed samples. To address this issue, a novel weed detection strategy was proposed in this study to identify weeds accurately in fields with varying coverage levels. For this purpose, red–green–blue (RGB) images of maize fields with different weed coverage levels were captured via a vertical take-off and landing fixed-wing unmanned aerial vehicle (UAV). The UAV images were first mosaicked, and a new weed detection strategy was developed and assessed. In this process, the MeanShift segmentation method, coupled with the local variance (LV) segmentation evaluation function and the Otsu automatic classification method, was initially employed to extract vegetation areas. The you-only-look-once (YOLO) v5n model was subsequently improved and used to detect maize plants. Finally, weed mapping was achieved by removing the identified maize plants from the vegetation through overlay analysis. The evaluation of the proposed method via an external dataset yielded favorable weed detection results, with an R 2 value of 0.96 and a root mean square error (RMSE) value of 3.08 % under the different weed coverage levels. Specifically, in addition to adjusting the activation function and the nonmaximum suppression method, the impacts of integrating various attention modules at different positions on the performance of the YOLO v5n model for maize plant detection were analyzed. Improving the YOLO v5n model by incorporating the efficient channel attention (ECA) module into the backbone of the original model and utilizing the Hardswish activation function is recommended. Overall, this study offers support for precise weed control. [Display omitted] • A new strategy was proposed for weed mapping in maize field. • Yolov5n model was improved by integrating attention module and altering activation function. • Weed mapping using proposed strategy has good consistency with the visually interpreted results. [ABSTRACT FROM AUTHOR]

• A hyperspectral data-processing framework was developed using Boruta and DNN. • Hyperspectral bands that can estimate the LNC of apple trees were explicitly identified. • Investigated the potential of phenological data to enhance apple LNC estimation. Rapid and accurate assessment of apple-tree leaf nitrogen content (LNC) plays an important role in smart orchards for accurate fertilization. Hyperspectral remote sensing technology provides a reliable, economical, and fast method to estimate LNC. However, high-dimensional spectral data contain redundant information that is irrelevant to the target parameters, necessitating the development of robust hyperspectral analysis techniques to estimate apple-tree LNC. Additionally, the role of phenological information in improving LNC estimates across multiple growth stages remains uncertain. The present study thus proposes a novel hyperspectral data-processing framework (Boruta-Iteration-DNN) to optimize estimates of apple-tree LNC and investigates the potential of phenological information to improve LNC estimates of multi-fertility apple trees. The proposed framework incorporates the Boruta feature-selection method for identifying important spectral bands and assessing how they affect model performance through iterative feature analysis. A deep neural network (DNN) regression model is then developed for LNC estimation. In addition, a widely used Pearson-DNN method is used for comparison to demonstrate the superiority of the proposed approach. Furthermore, the potential of incorporating phenological information (Day After Anthesis, DAA) into the estimation models is examined to evaluate its ability to better estimate LNC. The results demonstrate that the proposed framework not only enhances the accuracy of apple-tree LNC estimates with respect to the full-spectrum model (without DAA: validation R2 improved from 0.69 to 0.75, NRMSE reduced from 11.92 % to 9.97 %; with DAA: validation R2 improved from 0.72 to 0.79, NRMSE reduced from 11.09 % to 9.06 %), but also reduces the complexity and dimensionality of the models by eliminating 96 % of the redundant spectral bands (2015 bands). Moreover, the inferior results of the Pearson-Iteration-DNN framework further underscore the effectiveness of the Boruta method in selecting relevant spectral bands. Additionally, including the DAA phenological information enhances the accuracy of all models (the total R2 of training and validation increased by 0.05 ∼ 0.12, and the total NRMSE reduced by 1.49 % ∼ 1.93 %). These findings highlight the ability of DNN models integrated with hyperspectral data and phenological information to accurately assess LNC in apple trees, thereby offering valuable guidance to orchard management for developing precise fertilization strategies. [ABSTRACT FROM AUTHOR]

• We developed an RRCNet that fuses the deep and shallow spectral features to map RRC. • RRCNet was trained based on an RTM-based "soil-crop-crop residue" spectra dataset. • Use transfer learning (TL) to recalibrate the fully connected layer of RRCNet. • RRCNet + TL can provide more accurate RRC estimates than SI + statistical regression. Accurate determination of rice residue cover (RRC) can improve the monitoring of tillage information. Currently, the accurate determination of RRC using optical remote sensing is hindered by variations in cropland moisture and cover of following crops. The fractional cover (FC) of the soil (f S), crop (f C), and crop residue (f CR) changes (f S + f C + f CR = 1) after the following crop is planted, which increases the difficulty of remote-sensing RRC estimation. Cropland soil moisture and crop residue moisture affect the values of cropland and crop residue spectral indices (CRSIs), thereby reducing the accuracy of remote-sensing RRC estimation. Deep learning techniques (e.g., convolutional neural networks [CNN] and transfer learning [TL]) have been proven to extract the deep features of input images with distortion invariance, such as displacement and scaling, which are similar to moisture and the following crop effects on remote-sensing CRSIs. This study aimed to evaluate the combined use of deep features of cropland spectra extracted by deep learning techniques to estimate the cropland RRC under the effects of variations in cropland moisture and cover of the following crops. This study proposes an RRCNet CNN that fuses deep and shallow features to improve RRC estimation. A PROSAIL radiative transfer model was employed to simulate a cropland "soil–crop–crop residue" mixed spectra dataset (n = 103,068), considering the variations in cropland moisture and the cover of the following crop. The RRCNet was first pre-trained using the simulated dataset, and then the knowledge from the pre-trained RRCNet was updated based on field experimental FCs, RRCs, and Sentinel-2 image spectra using the TL technique. Our study indicates that RRCNet can incorporate shallow and deep spectral features of cropland "soil–crop–crop residue" mixed spectra, providing high-performance FCs and RRC estimation. The FCs and RRC estimates from RRCNet + TL (FCs: R 2 = 0.939, root mean squared error (RMSE) = 0.071; RRC: R 2 = 0.891, RMSE = 0.083) were more accurate than those from CRSI + multiple linear regression, CRSI + random forest, and CRSI + support vector machine (FCs: R 2 = 0.877–907, RMSE = 0.086–0.101; RRC: R 2 = 0.378–0.714, RMSE = 0.133–0.229). We mapped the multistage RRC based on Sentinel-2 multispectral instrument (MSI) images and RRCNet. Tillage information can be inferred from RRC and RRC difference maps changes. [ABSTRACT FROM AUTHOR]

• A-BiGRU enables more accurate modeling of crop growth to achiever high accuracy. • The Attention mechanism can focus on the critical growth period of the crop. • The bidirectional GRU structure is better than GRU. • Image reconstruction can effectively fill the data gap. Timely and accurate access to regional scale crop plant area and spatial distribution is essential for regional agricultural production and food security, especially in the context of global population growth and climate change. Deep learning has become prevalent in crop mapping under complex conditions due to its powerful feature extraction and nonlinear ability. This study proposes a time-series image classification framework using Attention-based Bidirectional Gated Recurrent Unit (A-BiGRU) to map rice, maize, and soybean in Fujin, China, from reconstructed Sentinel-2 time-series images. Firstly, the reconstructed Sentinel-2 time-series images with 10 spectral dimensions and 22 temporal dimensions were obtained by linear interpolation and Savitzky-Golay (SG) filter. Then, a neural network, the A-BiGRU was developed to identify different crops by taking advantage of their unique growth patterns. The attention mechanism enables temporal neural networks to focus on the critical growth periods of crops. Additionally, the structure of GRU is simpler than that of long short-term memory (LSTM) and simple recurrent neural network (SRNN), which reduces the number of parameters and alleviates overfitting. Compared to GRU, BiGRU can fully uses the time-series information of the entire crop growth cycle. To assess the effectiveness of the proposed method, we compared two deep learning methods (LSTM and SRNN) and three widely used non-deep learning classifiers (Spectral Angle Mapping (SAM), Support Vector Machine (SVM)) and eXtreme Gradient Boosting (XGBoost). The results demonstrate that A-BiGRU achieved the highest accuracy, with an overall accuracy of 0.9804, a macro F1 score of 0.9788 and a kappa score of 0.9714. We also selected four typical regions and compared the classification results with optical images, which showed that the proposed method has a good recognition effect. Therefore, the A-BiGRU method is capable of achieving high-precision crop mapping. [ABSTRACT FROM AUTHOR]

• The MP-Net model aggregates context information. • The MP-Net model reduces the feature losses. • The MP-Net model efficiently and precisely classifies crops. Accurate crop classification map is of great significance in various fields such as the survey of agricultural resource, the analysis of existing circumstance on land application, the yield estimation of crop and the disaster warning. The methods based on machine learning and deep learning are popularly used in crop classification and recognition of remote sensing images. However, the crop classification task based on neural networks still faces significant challenges due to the spatial and temporal distribution of crops and the inherent characteristics of remote sensing images. Therefore, this study proposes the multi-layer pyramid crop classification network (MP-Net) to solve the above problems. To reduce the feature loss during the crop extraction, the proposed model uses the pyramid pooling module to improve the ability of global information acquisition, and the information concatenation module to retain the upper features. Using the GF-6 and Sentinel-2 satellite data, the proposed model was tested in Erhai Lake Basin and Beian City. Compared with other five deep learning models, such as FCN, SegNet, U-Net, PSPNet and DeepLabv3+, the experimental results indicate that the proposed model achieves the highest accuracy in both study areas. Meanwhile, the proposed model has the advantages of short training time and high efficiency under the same running conditions. Overall, this study is beneficial to improve the efficiency and accuracy of crop classification task in the unbalanced temporal and spatial distribution. It also brings a feasible scheme for crop classification tasks in complex growing areas. The code has been publicly available at https://github.com/Xu-Chang-Hong/MP-Net. [ABSTRACT FROM AUTHOR]