Crop diseases and pests have a significant impact on planting costs and crop yields and, in severe cases, can threaten food security and farmers’ incomes. Currently, most researchers employ various deep learning methods, such as the YOLO series algorithms and U-Net and its variants, for the detection of agricultural plant diseases. However, the existing algorithms suffer from insufficient interpretability and are limited to linear modeling, which can lead to issues such as trust crises in current technologies, restricted applications and difficulties in tracing and correcting errors. To address these issues, a dual-module Kolmogorov–Arnold Network (U-PKAN) is proposed for agricultural plant disease detection in this paper. A KAN encoder–decoder structure is adopted to construct the network. To ensure the network fully extracts features, two different modules, namely Patchembed-KAN (P-KAN) and Decoder-KAN (D-KAN), are designed. To enhance the network’s feature fusion capability, a KAN-based symmetrical structure for skip connections is designed. The proposed method places learnable activation functions on weights, enabling it to achieve higher accuracy with fewer parameters. Moreover, it can reveal the compositional structure and variable dependencies of synthetic datasets through symbolic formulas, thus exhibiting excellent interpretability. A field corn disease image dataset was collected and constructed. Additionally, the performance of the U-PKAN model was verified using the open plant disease dataset PlantDoc and a gear pitting dataset. To better understand the performance differences between different methods, U-PKAN was compared with U-KAN, U-Net, AttUNet, and U-Net++ models for performance benchmarking. IoU and the Dice coefficient were chosen as evaluation metrics. The experimental results demonstrate that the proposed method achieves faster convergence and higher segmentation accuracy. Overall, the proposed method demonstrates outstanding performance in aspects such as function approximation, global perception, interpretability and computational efficiency.

Digital fringe projection profilometry often faces a trade-off between measurement accuracy and efficiency. Defocus technology is commonly employed to address this challenge and improve the efficiency of high-speed three-dimensional (3D) measurement. This technology uses 1-bit binary fringe patterns instead of traditional 8-bit sinusoidal patterns, but accurately measuring 3D shapes with both high speed and accuracy remains a challenge due to defocus errors. These errors are introduced by the manual adjustment of lens focal length and reduce both fringe pattern quality and measurement accuracy. To overcome this limitation, we propose a multi-stage generative adversarial network with a self-attention mechanism to correct inaccurate fringe patterns and transform them into more ideal sinusoidal fringe patterns. Our generation network comprises a multi-stage feature extraction network with a self-attention mechanism and an encoder-decoder network. A multi-stage network integrating residual and transformer modules is constructed to mine global feature information. The self-attention mechanism accurately detects key areas for correction, and the encoder-decoder network generates rectified sinusoidal fringe patterns by combining the feature information with the attention area. We use a discriminative network to evaluate whether the output of the generative network is good enough to be true. In our experiments, we considered different fringe widths and measured objects of various types and colors. The results show that our proposed method improves the quality of defocus fringe patterns and the accuracy of subsequent 3D reconstruction compared to existing direct defocus methods.

In transformer for semantic segmentation, patch embedding usually has only one convolutional layer with a large stride, leading to the decrease of feature extraction capability. In addition, the complex decoder results in high computation cost. To address the abovementioned two issues, we put forward a progressive downsampling transformer with convolution-based decoder (PDCDT), which is a simple, efficient yet powerful framework. Specifically, progressive downsampling layers for patch embedding are designed to refine the extracted features and reduce information loss at each stage of the hierarchical transformer encoder. Meanwhile, a simple decoder based on a convolution (conv) module is proposed for aggregating the characteristic information from multiscale (MS) output layers of the encoder, and it can realize dimensional transformation and information interaction with fewer parameters than the decoders used in the existing transformers. Extensive experiments show that PDCDT achieves competitive results on ADE20K mean intersection over union (47.9% mIoU) and cityscapes (82.6% mIoU). Finally, PDCDT is applied to gear pitting measurement in the gear contact fatigue test, and the comparative results indicate that PDCDT can improve the accuracy of pitting detection.

The judgment of gear failure is based on the pitting area ratio of gear. Traditional gear pitting calculation method mainly rely on manual visual inspection. This method is greatly affected by human factors, and is greatly affected by the working experience, training degree and fatigue degree of the detection personnel, so the detection results may be biased. The non-contact computer vision measurement can carry out non-destructive testing and monitoring under the working condition of the machine, and has high detection accuracy. To improve the measurement accuracy of gear pitting, a novel multi-scale splicing attention U-Net (MSSA U-Net) is explored in this study. An image splicing module is first proposed for concatenating the output feature maps of multiple convolutional layers into a splicing feature map with more semantic information. Then, an attention module is applied to select the key features of the splicing feature map. Given that MSSA U-Net adequately uses multi-scale semantic features, it has better segmentation performance on irregular small objects than U-Net and attention U-Net. On the basis of the designed visual detection platform and MSSA U-Net, a methodology for measuring the area ratio of gear pitting is proposed. With three datasets, experimental results show that MSSA U-Net is superior to existing typical image segmentation methods and can accurately segment different levels of pitting due to its strong segmentation ability. Therefore, the proposed methodology can be effectively applied in measuring the pitting area ratio and determining the level of gear pitting.

Gear pitting is a typical surface defect, and its accurate three-dimensional (3D) detection is significant for the operation and maintenance of equipment. Therefore, this paper develops a 3D gear pitting detection method based on digital twin. Firstly, a geometric model of gear pitting is built to expand the samples which are helpful improving the detection accuracy. Secondly, based on the gear pitting model and Unity, a virtual fringe projection profilometry (FPP) system is established in the metaverse for generating the deformation fringe patterns and retrieving their phases. To improve the accuracy of pitting detection, a gear pitting detection network (GPD-Net) is proposed. It can not only extract local and global characteristics from different spaces and scales but also fuse these features via the proposed attention-based conditional random field modules. Thus GPD-Net can retrieve the high-precision wrapped phase by a single fringe pattern. Meanwhile, an FPP system is built in the physical world, and the reconstructed actual point cloud images are imported into the virtual FPP system for the precise measurement of gear pitting. The developed measurement system can be regarded as a meta-defect-detection system. The experimental results show the superiority of GPD-Net over the state-of-the-art phase retrieval algorithms and the effectiveness of the proposed three-dimensional gear pitting detection method. The proposed 3D detection method based on FPP is not limited by the fault type, and it can be effectively applied to the 3D measurement of various surface defects.

Visual detection is an available approach for measuring gear pitting. Unfortunately, the number of gear pitting images is limited, resulting in that the detection accuracy of gear pitting is unsatisfactory. In order to augment gear pitting samples with different styles, a novel Cycle Generative Adversarial Network based on a symmetric tree structure (Tree-CycleGAN) is proposed. In Tree-CycleGAN, a new type of generator with tree structure named tree generator is designed to produce various types of high quality target samples from the source-domain samples, and a maximum diversity loss is constructed to enlarge the difference between two arbitrary branches; then a similar tree reconstructor is designed for translating target samples into source samples. Two discriminators are designed for making the generated images approximate to the target images in two cyclic processes. Via inception score, structural similarity indexes and peak-signal-to-noise ratio, the quality and diversity of images obtained by Tree-CycleGAN are evaluated. Comparative results show the superiority of Tree-CycleGAN over other domain adaptation GANs. The proposed Tree-CycleGAN combined with U-net has been successfully applied to gear pitting detection. Experimental results prove that the proposed methodology precedes the basic U-net method without sample augmentation and the method based on CycleGAN and U-net.

针对U-Net存在的小目标分割精度低、计算复杂度高、收敛慢的问题,构建基于空洞卷积和重构采样单元的U-Net网络(DSU-Net).在DSU-Net中,为增大图像特征提取的感受野并融合多尺度信息,设计具有不同膨胀率的空洞卷积层;针对池化过程丢失大量语义信息的缺点,构建将池化与卷积相结合的采样单元,并运用深度可分离卷积进行特征提取,从而增强神经网络的特征提取能力并降低计算成本.两个公开医学图像数据集的实验结果表明,在IoU、Dice Coeff和F1 Score三个评价指标上,DSU-Net较U-Net、ResU-Net、R2U-Net和U-Net++有着更好的分割性能.最后,将DSU-Net应用于齿轮点蚀的视觉测量,结果表明所提出方法能够更加精确地计算出齿轮点蚀面积率,从而解决了齿轮接触疲劳试验中高效准确检测齿轮失效的难题.

An efficient visual detection method is explored in this study to address the low accuracy and efficiency of manual detection for irregular gear pitting. The results of gear pitting detection are enhanced by embedding two attention modules into Deeplabv3 + to obtain an improved segmentation model called attention Deeplabv3. The attention mechanism of the proposed model endows the latter with an enhanced ability for feature representation of small and irregular objects and effectively improves the segmentation performance of Deeplabv3. The segmentation ability of attention Deeplabv3+ is verified by comparing its performance with those of other typical segmentation networks using two public datasets, namely, Cityscapes and Voc2012. The proposed model is subsequently applied to segment gear pitting and tooth surfaces simultaneously, and the pitting area ratio is calculated. Experimental results show that attention Deeplabv3 has higher segmentation performance and measurement accuracy compared with the existing classical models under the same computing speed. Thus, the proposed model is suitable for measuring various gear pittings.

Although convolutional neural networks (CNNs) have become the mainstream segmentation model, the locality of convolution makes them cannot well learn global and long-range semantic information. To further improve the performance of segmentation models, we propose u-shaped vision Transformer (UsViT), a model based on Transformer and convolution. Specifically, residual Transformer blocks are designed in the encoder of UsViT, which take advantages of residual network and Transformer backbone at the same time. What’s more, transpositions in each Transformer layer achieve the information interaction between spatial locations and feature channels, enhancing the capability of feature learning. In the decoder, for enhancing receptive filed, different dilation rates are introduced to each convolutional layer. In addition, residual connections are applied to make the information propagation smoother when training the model. We first verify the superiority of UsViT on Automatic Portrait Matting public dataset, which achieves 90.43% Acc、95.56% DSC and 94.66% IoU with relatively fewer parameters. Finally, UsViT is applied to gear pitting measurement in gear contact fatigue test, and the comparative results indicate that UsViT can improve the accuracy of pitting detection.

The traditional calculation methods of the pitting area ratio include artificial vision inspection and rubbing measurement based on cutting tooth. However, these methods have the disadvantages of low efficiency, high cost and static measurement. The non-contact computer vision measurement technology can achieve continuous monitoring without interfering with the machine operation, and have satisfactory detection accuracy. In this paper, we propose an integrated Yolov5-Deeplabv3 + real-time segmentation network (YDRSNet) for gear pitting measurement. The two-stage network is constructed by using Yolov5 and an improve Deeplabv3 + , which can be applied to process the video samples in real time and overcome the problem of sample imbalance. Considering that the second-stage network implements a binary classification task, the dice loss is applied to replace the Cross-entropy loss for reducing the amount of calculation and solving the problem of sample imbalance effectively. Moreover, a DC-Focus module is embedded into the second-stage network for reducing the information loss caused by down sampling. Compared with the existing typical segmentation algorithms, the proposed YDRSNet has stronger segmentation ability, and it can segment the effective tooth surface and different levels of pitting quickly and accurately. The proposed methodology provides a feasible way for online measuring the pitting area ratio and detecting the degree of gear failure.

Aiming at the problems of low segmentation accuracy of small targets, high computational complexity in U-Net, a new U-Net network based on dilated convolution and reconstructed sampling units (DSU-Net) is proposed. In DSU-Net, in order to increase the receptive field of image feature extraction and fuse multi-scale information, dilated convolutional layers with different dilation rates are designed; in view of the shortcoming of losing a large amount of semantic information during the pooling process, sampling units that combine pooling and convolution are constructed, and depthwise separable convolution is used for feature extraction, thereby enhancing the feature extraction capability of neural network and reducing the computational cost. The experimental results of Gear Pitting dataset show that DSU-Net has better segmentation performance than U-Net, ResU-Net and R2U-Net on the three metrics of IoU, Dice Coeff and F1 Score. The proposed method can calculate the gear pitting area ratio more accurately, so as to solve the problem of efficiently and accurately detecting gear failure in the gear contact fatigue test.

Since the traditional gear pitting detection technique based on artificial observation has low accuracy and efficiency, an accurate vision detection method based on a new development of Mask R-CNN is explored. In order to solve the issue of small pitting samples, the generated adversarial network is firstly employed to enlarge the training samples with multilevel pitting. Then, we put forward a multipath fusion Mask R-CNN with double attention (DAMF Mask R-CNN) to implement the simultaneous segmentation of tooth surface and gear pitting. This new type of Mask R-CNN enhances the ability of target segmentation by embedding two attention modules, which can strengthen the feature expression. Considering the difference in the grayscale texture shapes of different-level pitting images, a multicascade multipath feature extraction network is constructed by designing a multicascade multipath pyramid module and adding a dual attention module, which significantly improves the generalization ability and segmentation performance of small pitting. Finally, we segment the gear pitting and tooth surface by the DAMF Mask R-CNN neural networks, and then compare it with Mask R-CNNs, Mask Scoring R-CNNs, U-Net, and Deeplabv3+. It can be known from the comparative results that the proposed DAMF Mask R-CNN has higher detection accuracy and convergence performance than the traditional segmentation networks.

To accurately and quantitatively detect the gear pitting of different levels on the actual site, this paper studies a new vision measurement approach based on a tunable vision detection platform and the mask region-based convolutional neural network (Mask R-CNN). The shooting angle can be properly set according to the specification of the target gear. With the obtained sample set of 1500 gear pitting images, an optimized deep Mask R-CNN was designed for the quantitative measurement of gear pitting. The effective tooth surface and pitting was firstly and simultaneously recognized, then they were segmented to calculate the pitting area ratio. Considering three situations of multi-level pitting, multi-illumination, and multi-angle, several indexes were used to evaluate detection and segmentation results of deep Mask R-CNN. Experimental results show that the proposed method has higher measurement accuracy than the traditional method based on image processing, thus it has significant practical potential.

Stem diameter is an important parameter in the process of plant growth which can indicate the growth state and moisture content of the plant, its automatic detection is necessary. Traditional devices have many drawbacks that limit their practical uses in general case. To solve those problems, a stem diameter inspection spherical robot was developed in this study. The particular mechanism of the robot has turned out to be suitable for performing monitoring tasks in greenhouse mainly due to its spherical shape, small size, low weight and traction system that do not produce soil compacting or erosion. The mechanical structure and hardware architecture of the spherical robot were described, the algorithm based on binocular stereo vision was developed to measure the stem diameter of the plant. The effectiveness of the prototype robot was confirmed by field experiments in a tomato greenhouse. The results showed that the machine measurement data was linearly correlated with the manual measurement data with R2 of 0.9503. There was no significant difference for each attribute between machine measurement data and manual measurement data (sig > 0.05). The results showed that this method was feasible for nondestructive testing of the stem diameter of greenhouse plants. Keywords: stem diameter inspection, spherical robot, binocular stereo vision, Census transform DOI: 10.25165/j.ijabe.20191202.4163 Citation: Quan L Z, Chen C, Li Y J, Qiao Y J, Xi D J, Zhang T Y, et al. Design and test of stem diameter inspection spherical robot. Int J Agric & Biol Eng, 2019; 12(2): 141–151.

Based on the principle of directional growth and cultivation of maize,the present situation of directional planting mechanization at home and abroad is reviewed.According to current research status,the main problems in the development of maize directional planting mechanization in China are analyzed,and the key technologies of mechanized directional planting involving motion control and batch identification are investigated.The results show that maize directional planting technology has many advantages,such as high yield,disease prevention.At present,the disadvantages of the maize directional planting technology are more dependent on manual operation,which is time and energy consuming and causes difficulties in promoting large-scale mechanization;Researches on mechanized directional planting of maize is more inclined to the design of directional seeding mechanism,and is lack of the research on the basic theory of maize particle swarm directional alignment.In the light of the present situation of mechanized planting of maize,it is suggested that study on the mechanism of batch alignment of maize seeds should be carried out by means of multiple techniques.

The study of corn mechanization planting orientation is still in the primary stage. Design of relevant institutions is mostly based on conceptual design of ideal grain shape, and the study on the basic characteristics and basic theory on corn particle swarm oriented alignment technology is lack. The motion state is more complex and affects the oriented alignment effects of maize grain seriously in electromagnetic vibration modes on maize grain in line in the process of conveying and sorting grain. Understanding of mechanical properties and optimization of vibration source parameters are the premise and key to get better oriented alignment effect of maize grain in the study of maize grain in vibration field. At first, this paper established the model of maize grain sliding motion through the force analysis of corn grain in the electromagnetic vibration system, and also obtained the minimum amplitude of the system of 0.05 mm which was the critical condition of contacts between the kernel and platform. Secondly, considering the special physical characteristics of corn grain, 1000 grains for each kind among 5 different kinds of corn with high quality in directional cultivation conditions were selected, and the instruments were adopted, such as the electronic balance with precision electronic scale of 0.01 g, the BPG-9.30AH electric thermostat blast drying box and MXD-01 friction coefficient instrument. For the different types of maize grains, the measured water content is in the range of 13.88%-14.53%, the density range is 1.13-1.19 g/cm3, and the sliding friction coefficient range between corn and hard rubber is 0.58-0.79. The corn planting orientation still relies on artificial pendulum, lacking mechanized corn directional seeding device. In order to make corn group oriented alignment system achieve directional cultivation, directional seeding accuracy requirements, and design parameters of the electromagnetic vibration system for corn particle swarm oriented alignment performance, were analyzed, as well as kinematics and dynamics of the system. The relationship between seed oriented alignment process and design parameters of system was explored in order to satisfy critical conditions requirements; the selected parameters had significant influence on the working performance of the system, and the main parameters affecting the performance of oriented alignment were the direction angle of vibration, vibration amplitude and frequency. On the corn particle swarm oriented alignment test platform, through the 2 orthogonal rotation design tests with 3 factors, the design parameters were optimized, and the factor and test index (directional success rate, the success rate of the entire column) regression equation was established. The optimal design parameters range was as follows: Vibrating direction angle was 32°-35°, amplitude was 0.1-0.15 mm, and vibration frequency was 51.5-52 Hz; corresponding to the test index, directional success rate was in the range of 63.8%-67.4%, the success rate of entire column was 95.8%-96.3%; the optimal combination of design parameters was the vibration direction angle of 34.42°, the amplitude of 0.14 mm and the vibration frequency of 51.77 Hz. The best combination of parameters was used in system verification test, with the high speed camera system implementation of recording and analysis on grain oriented alignment process, and in the absence of mechanical assisted mechanism, the average success rate of the orientation reached 66.2%, the success rate of entire column was 97.4%. The research can provide reference for the directional planting of maize grain and the development of directional seedling-raising equipment.

In order to improve the mechanization level of corn sorting,a kind of mechanical device with high efficiency and batch transportation of maize grain was designed to solve the problem that traditional corn sorting machine of grain was difficult to realize the batch transportation.The main structure,working principle and related structure parameters of the batch transportation device were described.Structural design of key components of the device,analysis of performance characteristics of amplitude difference type electromagnetic vibration system and batch transportation track were introduced.EDEM virtual simulation test was done on the device performance of the batch transportation,track arc central angle,wedge baffle angle and track width were taken as experiment factors,and batch column performance indicators were selected as test index of the virtual orthogonal experiment.The influence of related factors on the performance of the batch transportation was analyzed,and reasonable parameter combination was obtained as:the track width of 15.4 mm,track arc central angle of 45° and wedge baffle angle of 110°.The bench test was carried out,the bench test results showed that the electromagnetic vibration amplitude was 1.5 mm,frequency was 52 Hz,the vibration direction angle was 28°,the installation angle was 0°,the added mass was 0.3 kg,one side of plate spring was adjusted to 87°,the other side of plate spring was adjusted to 35°,the amplitude of the track with added mass was 1 mm,the batch transportation qualified index was 83.1％,adhesion and retention index was 10.4％,hysteresis index was 6.5％.The batch transportation performance was good,which met the performance requirements of corn seed separator,and the research provided a theoretical and technical reference for the research and design of corn sorting device.

Field weeds have a great effect on the yield of normal economic crops. The commonly used methods of weeding include artificial weeding, mechanical weeding, weed quarantine, biological control, chemical control, and so on. Among those methods, chemical weeding is the best when it comes to sudden grass damage (the weeds of disaster break out suddenly), because it works well and takes less time. However, when we use chemical weeding, there are many problems such as low utilization rate of liquid medicine and serious pollution to the environment with overplus liquid. Therefore, toward-target pesticide application is proposed to solve these problems above all. As we all know, toward-target pesticide application machineries often use machine vision technology to obtain spatial location information of weeds, which makes the nozzle rigidly fixed on the machine and at the same time, the control system adjusts the valve to open and close to achieve the purpose of spraying. On this basis, we use spraying nozzle to change the direction of application, resulting in precision spraying in this paper. In order to study the law of motion for spray droplet, we established dynamic model based on application methods with the principle of aerodynamics, and then the formula of the droplet flying and the analytic formula of the falling point were deduced. Next, the numerical simulation analysis of the covering area was carried out. Especially, in the process of analysis, we discussed the resistance characteristics of spray droplets, the nozzle angle, nozzle speed, machine speed and injection pressure, which affect the movement process we discussed. What's more, the model established above can also be used to control the location and shape of the droplet cover area by adjusting the working parameters, to improve the accuracy of toward-target pesticide application. Furthermore, in order to verify the correctness of the model, we set up the toward-target pesticide application platform and selected the special test site. Under no wind conditions, we conducted a simple target spraying test indoor, and then measured the geometric characteristics of actual liquid dripping place carefully. At the same time, the relevant parameters were obtained by high-speed camera technology. Next, we took the data into the dynamic model mentioned above to get the theoretical coverage of the liquid medicine, which was then compared with the actual coverage of the liquid medicine. Finally, it can be found that there is an error between the calculated and experimental results, and the error range is 8%-13%. In addition, to study the sedimentary characteristics of liquid medicine with this application method, we analysed the distribution law of the deposition amount of liquid medicine in the effective coverage area by water absorption paper weighing method. According to the analysis, it can be seen that the peak value of sediment changes regularly with the parameters such as the change of the nozzle angle, which suggests that the physical background of the model is clear, and the correlation between the parameters of the model and the sprinkler type is small, so the model transplant is strong. In other words, it can be used as a theoretical basis for the relevant spraying machinery.

A new segmentation algorithm of touching corn kernels and a direction‐correcting method are proposed to separate touching corn kernels in digital image and realize the direction‐correcting processing of corn kernels ,based on the traditional wa‐tershed algorithm .Firstly ,in order to get the pretreatment image ,Wiener filter and mathematical morphology are adopted to re‐duce noises and clarify the background of the image .Then ,by combining with tow edge detection operators ,the boundary of touching corn kernels is determined in the image ,which can be served as watershed of the algorithm .Erode transform and extre‐mum treatments are used to construct the basin of the algorithm afterwards ,and then complete the watershed segmentation oper‐ation of the image using the watershed and basin .Finally ,the method of multi‐scale wavelet analysis is applied to correct the di‐rection of corn kernels .The results of experiment show that the correction rate of segmentation and direction‐correcting are 94%and 95 .6% by this algorithm .The algorithm effectively solves the problem of ‘over‐segmentation’ and‘leakage’ in the tradition‐al watershed segmentation algorithm and presents a great promotion than the traditional algorithm on efficiency and accuracy .

针对植物工厂狭小密闭工作空间(狭闭空间)内大范围作业需求,提出了一种应用于植物工厂内部的物流化搬运机器人.采用D-H法建立连杆坐标系,通过求解运动学正解对机器人运动空间进行分析,在此基础上,采用边界追踪法规划了一种满足植物工厂立体式培育特点的滑切式搬运最优轨迹,提取轨迹上若干点,求解提取点运动学逆解,采用三次样条插值拟合机械手臂关节变量随时间变化函数,并在实验室中搭建育秧环境可自动调控的立体式育秧平台进行试验,控制机器人按照最优轨迹运动,应用高速摄像技术对实际运动轨迹进行记录,试验结果表明,实际轨迹和最优轨迹最大绝对误差为8 mm,在误差允许范围内,可以完成植物工厂内所需的搬运作业功能.