Search

It is observed that banana production is affected by numerous diseases and causes a large loss to the poor farmers. By using modern image processing technology and deep learning methods, these diseases can be found at an earlier stage and appropriate precautions can be taken at time to avoid more injury and increase banana production. In this paper, a novel banana disease recognition method is proposed. First, a banana leaf feature enhancement framework is designed to enhance the banana features under the complex environment. Then, a novel neural network, EM-ERNet, is designed based on the Resnet backbone architecture. In the EM-ERNet, dilated convolution and multi-scale convolution are combined to improve the ability of feature extraction. Batch normalization is used to prevent the network over-fitting and enhance the model robustness. The ELM algorithm optimized by particle swarm (PELM) is used to speed up the fusion network in the final fully connected layer. The experiment was conducted on the public banana disease detection database. The highest detection accuracy for crown rot, venturia, scorch rot, anthrax and normal bananas could reach 94.27, 93.36, 89.39, 88.87 and 96.39% respectively, which was better than the existing related competitors. In addition, ResNet50 takes 80 min to identify, while EM-ERNet only takes 60 min. We can see that our method improves the parameters by a quarter.

Currently, deep learning technology is widely used in various fields, such as face recognition, object recognition, and image classification. Multiple application instances sharing GPU resources can deploy more applications with limited GPU resources. However, this will lead to resource competition problems, resulting in application switching, timeouts, and other phenomena. Therefore, how to deploy these applications to a single machine with limited resources and properly schedule tasks while maximizing system performance is a new challenge. In this article, we propose Tinker, a middleware, to solve multiple CNN-based application deployment problems on a single machine. Tinker has two phases: 1) offline analysis and 2) runtime scheduling. Offline analysis generates the best application deployment configuration information of the current system. Runtime scheduling can properly schedule tasks to ensure that they are completed normally and efficiently utilize resources to improve system performance. Our experiment proved that Tinker boosts system performance and ensures that most tasks are completed efficiently.

When detecting citrus surface defects, the performance of machine vision system is affected by different aspects such as the size, shape and environment. Therefore, a method for classifying citrus surface defects based on machine vision was proposed in this paper. First, the Fuzzy C-Means algorithm optimized by the Gray Wolf Optimizer algorithm was used to preprocess the citrus image. The citrus in the image was separated from the background; Then, the improved convolutional neural network combined with the State Transfer Algorithm (STA) was used to identify the citrus surface defects. We selected 2000 Tribute Citrus, 1000 ones with and without the defects separately, to carry on the experiment. The identification accuracy of the trained model on the dataset was 99.1%. In order to verify the effectiveness of the model in complex background, the convolutional neural network in combination with a STA was compared with SVM, AlexNet, VGG16 and other methods. The experimental results show that the citrus surface defect classification method based on machine vision is effective.

The identification of maize leaf diseases will meet great challenges because of the difficulties in extracting lesion features from the constant-changing environment, uneven illumination reflection of the incident light source and many other factors. In this paper, a novel maize leaf disease recognition method is proposed. In this method, we first designed a maize leaf feature enhancement framework with the capability of enhancing the features of maize under the complex environment. Then a novel neural network is designed based on backbone Alexnet architecture, named DMS-Robust Alexnet. In the DMS-Robust Alexnet, dilated convolution and multi-scale convolution are combined to improve the capability of feature extraction. Batch normalization is performed to prevent network over-fitting while enhancing the robustness of the model. PRelu activation function and Adabound optimizer are employed to improve both convergence and accuracy. In experiments, it is validated from different perspectives that the maize leaf disease feature enhancement algorithm is conducive to improving the capability of the DMS-Robust Alexnet identification. Our method demonstrates strong robustness for maize disease images collected in the natural environment, providing a reference for the intelligent diagnosis of other plant leaf diseases.

Background: Under natural light irradiation, there are significant challenges in the identification of maize leaf diseases because of the difficulties in extracting lesion features from constantly changing environments, uneven illumination reflection of the incident light source and many other factors.Results: In the present paper, a novel maize image recognition method was proposed. Firstly, an image enhancement framework of the maize leaf was designed, and a multi-scale image enhancement algorithm with color restoration was established to enhance the characteristics of the maize leaf in a complex environment and to solve the problems of high noise and blur of maize images. Subsequently, an OSCRNet maize leaf recognition network model based on the traditional ResNet backbone architecture was designed. In the OSCRNet maize leaf recognition network model, an octave convolution with characteristics to accelerate network training was adopted, reducing unnecessary redundant spatial information in the maize leaf images. Additionally, a self-calibrated convolution with multi-scale features was employed to realize the interactions of different feature information in the maize leaf images, enhance feature extraction, and solve the problems of similarity of maize disease features and easy learning disorders. Concurrently, batch normalization was employed to prevent network overfitting and enhance the robustness of the model. The experiment was conducted on the maize leaf image data set. The highest identification accuracy of rust, grey leaf disease, northern fusarium wilt, and healthy maize was 94.67%, 92.34%, 89.31% and 96.63%, respectively. Conclusions: The aforementioned methods were beneficial in solving the problems of slow efficiency, low accuracy and image recognition training, and also outperformed other comparison models. The present method demonstrates strong robustness for maize disease images collected in the natural environment, providing a reference for the intelligent diagnosis of other plant leaf diseases.

Aiming at solving the problems of high background complexity of some butterfly images and the difficulty in identifying them caused by their small inter-class variance, we propose a new fine-grained butterfly classification architecture, called Network based on Multi-rate Dilated Attention Mechanism and Multi-granularity Feature Sharer (MRDA-MGFSNet). First, in this network, in order to effectively identify similar patterns between butterflies and suppress the information that is similar to the butterfly’s features in the background but is invalid, a Multi-rate Dilated Attention Mechanism (MRDA) with a symmetrical structure which assigns different weights to channel and spatial features is designed. Second, fusing the multi-scale receptive field module with the depthwise separable convolution module, a Multi-granularity Feature Sharer (MGFS), which can better solve the recognition problem of a small inter-class variance and reduce the increase in parameters caused by multi-scale receptive fields, is proposed. In order to verify the feasibility and effectiveness of the model in a complex environment, compared with the existing methods, our proposed method obtained a mAP of 96.64%, and an F1 value of 95.44%, which showed that the method proposed in this paper has a good effect on the fine-grained classification of butterflies.

Space-based detection satellite and remote detection phased array radar are important parts of space security detection system. The aircraft’s detection, tracking and parameter estimation, trajectory prediction with detection satellite and radar are primary problems to be solved in the early detection of ballistic aircraft. This paper mainly studies the simulation of aircraft’s trajectory in early detection phase, the establishment of a dynamic model of the active segment and trajectory simulation and the simulation to generate a multi-level trajectory vehicle’s trajectory data based on the estimation of the key parameters of the aircraft in the single star observing conditions, the trajectory forecast and radar observation conditions for aircraft tracking. Due to the incompleteness of measurements for the single-satellite detection and the bad convergence, this paper proposes a fired at the focal plane method based on the priori template, establish a complete formula derivation algorithm processes, establish a priori standard ballistic template with the simulation trajectory data and the validity of the method using Monte Carlo simulation. Based on the MATLAB graphical user interface, it builds a simulation platform of ballistics aircraft detection probe which can effectively complete the early detection of scene simulation and demonstration. The simulation results show that the method can solve the bad convergence problems of the detection of a single star and it suits for the application to the ballistic vehicle’s key point estimation.

In this paper, the state of charge (SOC) is estimated by charging the battery at constant current and voltage, collecting the current, voltage, temperature and internal resistance during the experiment. The influence of internal resistance on SOC prediction of lithium batteries is mainly considered. In this paper, the Improved BP neural network is used to carry out simulation experiments. The experimental results show that the prediction accuracy is higher and the simulation effect is better than that without considering the internal resistance.

Different from homogeneous clusters, when distributed training is performed in heterogeneous clusters, there will be great performance degradation due to the effect of stragglers. Instead of the synchronous stochastic optimization commonly used in homogeneous clusters, we choose an asynchronous approach, which does not require waiting for stragglers but has the problem of using stale parameters. To solve this problem, we design a straggler-aware parameter server (SaPS), which can detect stragglers through the version of parameters and mitigate their effect by a coordinator which can limit the staleness of parameters without waiting for stragglers. Experimental results show that SaPS can converge faster than fully synchronous, fully asynchronous and some SGD variants.

Ultrasonically-assisted drilling (UAD) is increasingly used in orthopedic surgery. Studies have shown that this method has certain advantages in reducing cutting force, improving surface quality and reducing tool wear. The purpose of this study was to use genetic algorithm (GA) to find the best combination parameters (cutting parameters: feed rate and spindle speed, ultrasonic parameters: frequency and amplitude). First this study builds a platform and central composite design was employed to carry out the bone drilling test. Based on the results, a response surface model was developed, which was used as a fitness function for GA. The investigation showed that the technique of can effectively estimate the optimal setting of bone drilling parameters and ultrasonic vibration parameters for minimum thrust force value, which can provide some guiding significance for clinical medicine to a certain extent.

In the existing machine vision technology for tomato leaf disease recognition, due to interference from the external environment, it is easy to generate noise during image acquisition, and the characteristics of different diseases are similar, which makes image disease recognition difficult. Therefore, we propose a new framework for tomato leaf disease recognition. First, the image is denoised and enhanced by Binary Wavelet Transform combined with Retinex (BWTR), noise points and edge points are removed, and important texture information is retained. Then, the tomato leaves were separated from the background using KSW optimizatied by Artificial Bee Colony algorithm (ABCK). Finally, the Both-channel Residual Attention Network model (B-ARNet) was used to identify the pictures. The application results of 8616 images show that the overall detection accuracy is about 89%. Experiments show that the tomato leaf disease recognition method based on the combination of ABCK-BWTR and B-ARNet is effective.

Aiming to solve many people do not know how to use computers to solve practical problems in life and work, we design a series of detailed steps to analyze, design, represent, program and verify with computing thinking. We demonstrate our steps in detail with an example to help them which will encounter similar problems in the future.

In view of the current situation that the experimental teaching hours are few and the task is heavy, the offline progressive experimental teaching, online micro class teaching and online progressive practical training breakthrough teaching mode are adopted, the experimental teaching contents with progressive logical relationship are designed, the difficult and in-depth practical training tasks are allocated, and the trainees’ practical training results are mastered openly, fairly and fairly in real time on the training platform It is proved that this method can improve the quality of practical teaching.

Based on wireless sensor network, a system-level solution of wireless sensor network for oil well safety information collection is constructed, including hardware and software design of data acquisition, data transmission and data analysis and processing. The system has the characteristics of low cost, wide range and flexible layout, which solves the problems of cable networking, such as wiring, location and slow troubleshooting problem. The experiment shows that the system meets the actual requirement and runs stability.

This paper describes the innovative design of a three-dimensional (3D) motion device based on a flexible mechanism, which is used primarily to produce accurate and fast micro-displacement. For example, the rapid contact and separation of the tool and the workpiece are realized by the operation of the 3D motion device in the machining process. This paper mainly concerns the device performance. A theoretical model for the static performance of the device was established using the matrix-based compliance modeling (MCM) method, and the static characteristics of the device were numerically simulated by finite element analysis (FEA). The Lagrangian principle and the finite element analysis method for device dynamics are used for prediction to obtain the natural frequency of the device. Under no-load conditions, the dynamic response performance and linear motion performance of the three directions were tested and analyzed with different input signals, and three sets of vibration trajectories were obtained. Finally, the scratching experiment was carried out. The detection of the workpiece reveals a pronounced periodic texture on the surface, which verifies that the vibration device can generate an ideal 3D vibration trajectory.

In-air gesture recognition is becoming a popular means of interacting with terminal devices, such as smart phones, tablets and laptops. To implement in-air gesture recognition, many methods have been researched, such as computer visionbased, RF-based, WiFi-based and ultrasonic-based. Compared to other approaches, the ultrasonic-based method shows superiorities in low-cost, robustness and so on. In this paper, we construct a dual-frequency based ultrasonic gesture recognition system called DopGest, which utilizes the Doppler Effect to recognize gestures. Our system is implemented completely on a laptop without any additional equipments, and we use two speakers and one microphone already embedded in laptop as sensors. Most of ultrasonic-based methods exploit single tone, which limit the type of gestures. For instance, they can not recognize horizontal swipe gestures. On the contrary, our system can recognize swipe left and swipe right gestures by using two high tones (18 kHz). In our system, DopGest generates two different high frequency tones from two speakers and detects Doppler Effect caused by hand movements, then it extracts features from Doppler shifts of these two high tones, since laptops generally contain at least two speakers. Compared to computervision based method, our system can work under any light conditions without extra hardware devices and with low power consumption. Experiments show that our system can achieve an average classification accuracy of 88.9%.

This paper investigates the video delivery on the underlay device-to-device (D2D) communications. Considering that the distributed scheme requires less signaling overhead than the centralized one, we exploit Stackelberg game to establish a QoE-driven distributed cross-layer video delivery model to optimize the parameters which include the RB-level transmission power at the physical layer, a group of RBs at the data link layer and source coding rate at the application layer. In the model, the evolved NodeB (eNB), as a leader, plays an important role in guiding D2D pairs to reuse wireless resources intelligently. D2D pairs, as followers, achieve the goal of maximizing their individual quality of experience (QoE) by competing selfishly in a non-cooperative game. We prove the existence of the Stackelberg equilibrium for the established Stackelberg game model in mathematics and propose a QoE-driven Distributed Cross-layer Video-delivery (QDCV) scheme to reach the Stackelberg equilibrium. Extensive simulations show that the proposed scheme can achieve the Stackelberg equilibrium of the whole system in a relatively short time and our proposed scheme improves the average MOS by about twice compared with the existing QoE scheme.

This paper is concerned with the problem of joint resource allocation and user-pairing in virtual MIMO SC-FDMA systems to improve service quality of experience (QoE). No-reference logarithmic model is introduced to quantify service experience for each user and the objective is to maximize sum of all user’s mean of score (MOS). We firstly formulate the optimal problem into an S-dimensional (S-D) assignment problem. Then, to solve this problem, the modified Lagrangian relaxation algorithm is deduced to obtain the suboptimal result of joint user-paring and subchannel allocation. The merits of this solution are as follows. First, the gap between its results and the global optimal one can be quantified and controlled by balancing the complexity and accuracy, which merit the other suboptimal algorithms do not have. Secondly, it has the polynomial computational complexity and the worst case complexity is O(3LN³), where L is the maximum iteration time and N is the number of subchannels. Simulations also prove that our proposed algorithm can effectively improve quality of experience and the gap between our proposed and the optimal algorithms can be controlled below 8%.

The vigorous growth of big data has triggered both opportunities and challenges in business and industry. However, Web big data distributed in diverse sources with multiple data structures frequently conflict with each other, i.e. inconsistency in cross-source Web big data. In this paper, we propose a state-of-the-art architecture of auto-discovering inconsistency with Web big data. Our contributions include: (1) we classify the inconsistency features to formalize inconsistency data and establish an algebraic operation system, (2) we propose three algorithms to auto-discover inconsistency, including constraint-based, SDA-based and HPDM-based method and (3) we conduct experiments on real-world dataset to compare aforesaid schemes with Oracle-based inconsistency detection framework. The empirical results show that our methods outperform traditional framework both on accuracy and efficiency under Web big data.

In this paper, we present a multi-objective Web Services selection algorithm, which aims to achieve the dynamic web service composition. Firstly, we analyze multi-objective optimization problems with user constraints. Furthermore, we optimize the different QoS parameters of the workflow. And experimental results prove our algorithm is effective for solving the Web Services selection problem.

Identification of the same commodity entities is a major challenge in the heterogeneous multi-source e-commerce of big data. This paper introduces a framework based on Map-Reduce, called IIRS, which is made up of data index, data integration, entity recognition and data sorting. IIRS aims to form the unified model and high efficient commodity information with building an index model based on commodity’s attribute/value and constructing a global model map to record commodity’s attribute and value, identify the commodity entities in different e-commerce with measuring the similarity of the commodity’s identity, and then output the same identity commodity sets and their associated properties organized in the inverted index list. Through an extensive experimental study on real e-commerce dataset on Hadoop, IIRS significantly demonstrates its feasibility, accuracy, and high efficiency.

Concurrent Multipath Transfer (CMT) using multiple paths for simultaneous parallel transmission is a promising technology in next generation mobile networks, especially for bandwidth-hungry services. Due to the multi-homing feature of Stream Control Transmission Protocol (SCTP), researchers proposed to extend it to support CMT in order to distribute packets within the same flow over independent homogeneous paths. However, CMT over heterogeneous paths may cause the packet “reordering” problem in the receiver buffer and deteriorate the data delivery efficiency. To address this issue, we propose an Adaptive Delay-based Packet Splitting Mechanism (ADPSM) for CMT over heterogeneous paths. ADPSM defines an adaptive variable called sliding delay threshold to pick out the candidate set from all heterogeneous paths and distribute new arriving packets to the target path with minimum delay. Simulation results show that ADPSM mechanism is better than original CMT-SCTP with respect to the in-order packets arrival and average throughput, especially in the case of a low receiver buffer size.

Moving target tracking is on the basis of two-dimensional image processing technology, its main task is to use the projection of moving object was carried out on the field of moving target in image sequence in real time observation, and the observed object classification, and then analyze their trajectory. And target tracking based on image sequence, is through the sensor captured image sequence were analyzed, and then detect the target in the independence movement in the image, the region of interest or user, then estimate the target area in the image in each frame in the position. Usually, the target track is based on image sequence, but also have based on compressed video stream. Based on this theory, we propose a method to get the best tracking performance.

Based on the ontology construction thinking, knowledge base of C programming language is presented. Ontology construction principles, methods and advantages are analyzed, and factors such as the characteristics of knowledge structure and C programming language design course are also taken into consideration. Construction methods and progress of C programming language are detailed. By this way, the ontology C programming language course knowledge points are linked together, and its structure is also clear, to strengthen the teaching resources management and improve the teachers' and students' curriculum resources sharing and reuse.

In wireless communication systems, the combination of multiple-input multiple-output (MIMO) with orthogonal frequency division multiplexing (OFDM) is regarded as one promising techniques to support high data rate and high performance. In such a situation, wireless multimedia services have spread widely over MIMO-OFDM systems. How to provide user-satisfied multimedia services over MIMO-OFDM systems has attracted increasingly attentions. Quality of experience (QoE) is the key criteria for evaluating users' satisfaction with multimedia services in terms of mean opinion score (MOS). In this paper, we propose a QoE-driven cross-layer optimized video transmission scheme over MIMO-OFDM systems. No-reference QoE estimation model is applied to acquire the expected MOS taking account of video content type (CT), frame rate (FR), video sending bit rate (SBR) and packet error rate (PER). The aim of our proposed scheme is to achieve user-satisfied video quality by jointly optimizing video coding parameters at the application layer and modulation and coding scheme (MCS) at the physical layer. The problem is formulated to find the optimal combination of parameters to maximize expected MOS with constraint of maximum transmission delay, and solved by improved traversal algorithm. Experimental results have shown significant performance enhancement of the proposed system based on H.264/AVC and NS2.

How to provide user-satisfied video services must be solved in the future wireless communication systems. However, most of existing works focused on the PHY-MAC cross layer schemes do not consider the end-to-end video quality of service (QoS) and hence can only provide limited user-perceived video quality improvements. In this paper, a novel video transmission scheme with guaranteed end-to-end QoS over Multiple Input Multiple Output and Orthogonal Frequency Division Multiplexing systems is proposed. The end-to-end video quality is measured by recursive optimal pixel estimation (ROPE) algorithm. Our objective is to minimize the ROPE distortion by jointly optimizing the source encoding parameters of H.264/AVC encoder in the application layer, the adaptive modulation and coding parameters and time-frequency resource block selection in the physical layer. The experimental results demonstrate that our proposed scheme can provide good user perceived quality.

In wireless video streaming, the packet loss often occurs and affects the end-user visual quality. To alleviate the transmission error effects, intra refresh coding is usually used to improve the streaming error resilience ability from the view of source coding. At the physical layer, the adaptive modulation and coding (AMC) is also used to promote the transmission reliability at the transporting level. Both the error control components have their own influences on the received video quality. To achieve the best video transmission performance, it is crucial to make an error control tradeoff between intra refresh coding and AMC. In this paper, we propose an end-to-end video distortion optimized cross-layer error control method which jointly considers the video quantization parameter (QP) and intra refresh rate at the application layer, and AMC at the physical layer for delay-constraint real-time video streaming. The experimental results show that the proposed cross-layer error control streaming method can achieve the superior objective and subjective performances to the layer-independent error control streaming methods with and without cross-layer optimization.

In this paper, channel adaptive resource allocation and user scheduling problem for cognitive radio multiple-inputmultiple-output orthogonal frequency-division multiple access systems is investigated. Furthermore, stochastic approximation is employed to develop an on-line algorithm which is capable of adaptively learning the underlying channel distribution. The NP-hard combinatorial resource allocation and user scheduling problem is then solved through via the Lagrangian relaxation approach. Numerical results demonstrate the effectiveness of our proposed scheme.

Optimal power control for relay-assisted cognitive radio networks in which secondary users are assisted by cognitive relays is investigated. A monomial approximation based distributed geometric programming algorithm (MaDGP) is developed to maximize the secondary system utility under the individual and total power constraints of secondary users as well as the interference temperature limits of the primary users. The presented power control algorithm is distributively formulated using geometric programming and is then solved using the Lagrangian primal-dual decomposition. Numerical results demonstrate the effectiveness of our proposed algorithm.

Optimal power control for relay-assisted cognitive radio networks in which secondary users are assisted by cognitive relays is investigated. A monomial approximation based power control algorithm (MaPC) is developed to 1) maximize the secondary system weighted sum rates; 2) minimize the secondary system weighted sum powers under the individual and/or total power constraints of secondary users as well as the interference temperature limits of the primary users. The presented power control schemes are formulated using geometric programming and are then converted into equivalent convex optimization problems. Numerical results demonstrate the effectiveness of our proposed schemes.

License Plate Recognition (LPR) system is an important part of Intelligent Traffic System (ITS), which is very valuable in practical applications such as public safety, traffic management and military department. The license plate image edge detection is a hotspot to the process of license plate location and recognition. The edge detection affects the follow-up quality of the location and identification. This paper describes the various edge detection algorithms and operators. It gives the general steps of edge detection and analyses the existing difficulties. Through the simulation of the edge image on the plate, it analyzes the feasibility of the algorithm and summarizes the advantages and disadvantages of five detection operator which provides a theoretical basis for the next license plate location and identification.v

Video plus depth based 3D video gradually goes mobile. Taking into account the time-varying characteristics of wireless communication, this paper proposes a multi-pass varying bit-rate (VBR) control method for mobile 3D video coding. With the collected rate-distortion (RD) information and the reconstructed compressed video and depth in the first and second pass offline encodings, the encoder can establish the virtual view quality model (VVQM) to assess the right virtual view qualities under different rate combinations of video and depth. In the third pass encoding, the encoder utilizes VVQM to find the optimal video/depth target rate allocation, and then independently controls the video and depth rates with the rate-quantization (RQ) model. In the course of rate control, the video/depth rate allocation will be adjusted in real time to meet the varying channel bandwidth constraint. Experimental results show that the proposed rate control method can not only satisfy the VBR transmission requirements of mobile 3D video, but also provide the higher virtual view quality than the rate control method with fixed video/depth ratio rate allocation.

In this paper, we have formulated the problem of cross layer throughput-maximized dynamic resource management with guaranteed QoS in MIMO OFDM beamforming systems. In order to reduce of the amount of feedback information and computation complexity, OFDM subcarriers are grouped into subchannels based on the fact that adjacent subcarriers experience the same fading. According to characteristics of various traffics in actual networks, four types of traffics are classified. Different class has different QoS requirements and one connection only belongs to one class. Considering MAC QoS requirements, PHY channel side information (CSI) and connection fairness, the scheduling priority factor (SPF) is defined for each connection. Based on the scheduling priority factor, the suboptimal solutions for the formulated problem are obtained. Packets from different connections are queued at the base station. Scheduling and subchannel allocation are operated at jointly. Then, power allocation and bit loading is made for each packet. The performance of our proposed algorithm is also studied by simulation.

In this paper, we have derived the form of the maximum capacity for MIMO-OFDM system by adaptively allocating power in space, time and frequency domains in the case of perfect channel side information at the receiver (CSIR) and a zero-mean spatially white (ZMSW) channel distribution at the transmitter. Both the effects of Rayleigh fast fading and shadowing on the ergodic capacity are considered. The ergodic capacities achieved by the three different power allocation schemes, filling water in space-time-frequency domain, filling water in space-frequency domain and equally allocating power, are compared. For the ergodic capacity, the optimal power allocation scheme is filling water in space-time-frequency domain, which can obtain the maximal capacity as well as reduce the computation complexity.