A Review on the Application of Deep Learning Methods in Detection and Identification of Rice Diseases and Pests

In rice production,the prevention and management of pests and diseases have always received special attention.Traditional methods require human experts,which is costly and time-consuming.Due to the complexity of the structure of rice diseases and pests,quickly and reliably recognizing and locating them is difficult.Recently,deep learning technology has been employed to detect and identify rice diseases and pests.This paper introduces common publicly available datasets;summarizes the applications on rice diseases and pests from the aspects of image recognition,object detection,image segmentation,attention mechanism,and few-shot learning methods according to the network structure differences;and compares the performances of existing studies.Finally,the current issues and challenges are explored fromthe perspective of data acquisition,data processing,and application,providing possible solutions and suggestions.This study aims to review various DL models and provide improved insight into DL techniques and their cutting-edge progress in the prevention and management of rice diseases and pests.

Survey and Detection of Disease and Pest Occurrence in Citrus Nurseries

[Objectives]The paper was to ascertain the prevalence of diseases and pests in a range of citrus nurseries situated in Guangdong Province and its neighboring provinces.[Methods]Citrus diseases and pests were systematically investigated,and citrus leaf samples were randomly collected from 15 citrus nurseries across 8 regions in Guangdong Province and its neighboring provinces.Quantitative polymerase chain reaction(qPCR)and reverse transcription polymerase chain reaction(RT-PCR)techniques were employed to detect diseases in the collected samples.Additionally,root and substrate samples were obtained,and root-knot nematodes were isolated using the Baermann funnel method.[Results]The positive detection rate of citrus huanglongbing(HLB)was recorded at 3%,indicating an increase in attention towards this disease compared to 2013.Additionally,the positive detection rate for citrus bacterial canker disease(CBCD)was found to be 16.5%.It was observed that the majority of nurseries with positive samples employed open field rearing practices without the use of mesh chambers,and the primary source of scions was self-propagation.The detection rate of citrus tristeza virus(CTV)was found to be the highest,with a positive detection rate of 63%,and the prevalence in disease-bearing nurseries reached as high as 90%.In comparison to 2013,there had been no improvement in the condition of seedlings affected by CTV.The positive detection rate of citrus yellow vein clearing virus(CYVCV)was found to be 38%,with 70%of the surveyed nurseries exhibiting the disease.The citrus varieties identified as carriers of the disease included‘Qicheng’,‘Shatangju’,‘Wogan’,and‘Gonggan’.Nematodes were isolated from the matrix and roots of seedlings grown in both container and open field environments.The susceptibility of container seedlings to nematodes was found to be 36.4%,while the susceptibility of open field seedlings was 38.6%.Statistical analysis indicated no significant difference in susceptibility between the two groups.[Conclusions]The disease detection rates associated with various seedling rearing methods and citrus varieties exhibited notable variability.Open field seedlings without the protection of mesh chambers demonstrated a higher susceptibility to disease.Additionally,the types of infectious diseases varied among the different citrus varieties.

Establishment of High-sensitivity Rapid Fluorescence Quantitative Detection Method for Antibody against Peste des Petits Ruminants Virus

[Objectives]This study was conducted to establish a rapid quantitative method for detecting antibody against Peste des Petits Ruminants Virus(PPR V)in sheep serum.[Methods]Soluble N protein and NH fusion protein were obtained in Escherichia coli prokaryotic expression system by optimizing codons and expression conditions of E.coli.Furthermore,based on the purified soluble N protein and NH fusion protein,a high-sensitivity fluorescence immunoassay kit for detecting the antibody against PPR V was established.[Results]The method could quickly and quantitatively detect PPR V antibody in sheep serum,with high sensitivity and specificity,without any cross reaction to other related sheep pathogens.The intra-batch and inter-batch coefficients of variation were less than 10%and 15%,respectively,and the method had good repeatability.Through detection on 292 clinical serum samples,it was compared with the French IDVET competitive ELISA kit,and the coincidence rate of the two methods reached 93.84%.Compared with the serum neutralization test,the detected titer value of the high-sensitivity rapid fluorescence quantitative detection method was basically consistent with the tilter value obtained by the neutralization test on the standard positive serum(provided by the WOAH Brucellosis Reference Laboratory of France).[Conclusions]This method can realize rapid quantitative detection of PPR V antibody on site,and has high practical value and popularization value.

Mango Pest Detection Using Entropy-ELM with Whale Optimization Algorithm

Image processing,agricultural production,andfield monitoring are essential studies in the researchfield.Plant diseases have an impact on agricultural production and quality.Agricultural disease detection at a preliminary phase reduces economic losses and improves the quality of crops.Manually identifying the agricultural pests is usually evident in plants;also,it takes more time and is an expensive technique.A drone system has been developed to gather photographs over enormous regions such as farm areas and plantations.An atmosphere generates vast amounts of data as it is monitored closely;the evaluation of this big data would increase the production of agricultural production.This paper aims to identify pests in mango trees such as hoppers,mealybugs,inflorescence midges,fruitflies,and stem borers.Because of the massive volumes of large-scale high-dimensional big data collected,it is necessary to reduce the dimensionality of the input for classify-ing images.The community-based cumulative algorithm was used to classify the pests in the existing system.The proposed method uses the Entropy-ELM method with Whale Optimization to improve the classification in detecting pests in agricul-ture.The Entropy-ELM method with the Whale Optimization Algorithm(WOA)is used for feature selection,enhancing mango pests’classification accuracy.Support Vector Machines(SVMs)are especially effective for classifying while users get var-ious classes in which they are interested.They are created as suitable classifiers to categorize any dataset in Big Data effectively.The proposed Entropy-ELM-WOA is more capable compared to the existing systems.

Method for pests detecting in stored grain based on spectral residual saliency edge detection

Pests detecting is an important research subject in grain storage field.In the past decades,many edge detection methods have been applied to the edge detection of stored grain pests.Although some of them can realize the stored grain pests detecting,precision and robustness are not good enough.Spectral residual(SR)saliency edge detection defines the logarithmic spectrumof image as novelty part of the image information.The remaining spectrumis converted to the airspace to obtain edge detection results.SR algorithm is completely based on frequency domain processing.It not only can effectively simplify the target detection algorithm,but also can improve the effectiveness of target recognition.The experimental results show that the edge results of stored grain pests detected by SR method are effective and stable.

Pest Detection Method Using Multi-Fractal Analysis

A novel method for pest detection is proposed based on the theory of multi-fractal spectrum to extract pests on plant leaves.Both local and global information of the image regularity were obtained by multi-fractal analysis.By applying fractal dimension,the spots on leaf images can be extracted without loosing or introducing any information.The different parts of images are re-analysis by morphology operations to determine the candidate regions of pests.The performance of multi-fractal analysis of whitefly detection is investigated through greenhouse experiments.The experimental results show that the proposed method is robust to noise from light and is not sensitive to the complex environment.

多特征融合方法在马铃薯图像快速检测中的应用

针对目前马铃薯图像病虫害检测方法识别精度低、效率不高等问题。这里提出了一种主成分分析的特征加权融合自适应算法和改进支持向量机相结合的马铃薯病虫害快速检测方法。主成分分析的特征加权融合自适应算法完成特征块选择、主分量提取、加权和融合。融合后的特征采用决策树的思想,通过改进的支持向量机逐级分类。通过试验进行对比分析。结果表明,与传统的检测方法相比,该方法具有更高的检测精度和更短的执行时间。在病虫害、3类病害和10类虫害的检测准确率分别为98.45%、97.33%和98.00%,运行时间分别为4.81s、3.74s和4.65s。该检测方法为图像病虫害快速检测技术的发展提供了理论方法和依据。

基于改进YOLOv8的轻量化荷叶病虫害检测模型

腐败病、叶斑病、病毒病、斜纹夜蛾等荷叶病虫害严重影响莲子的产量与品质。开展疫病叶片检测是防治荷叶病虫害的重要措施。该研究以提高对荷叶病虫害的检测精度、减少模型的计算规模、提升可部署性为目标,提出了一种基于改进YOLOv8的轻量化荷叶病虫害检测模型,同时,建立了一种考虑不同环境条件的荷叶病虫害数据集。首先,将YOLOv8颈部网络中的卷积模块(Conv)替换为GSConv,将C2f模块替换为VoV-GSCSP,形成了Slim-neck架构,使模型在保持较高识别准确性的基础上降低计算复杂度。同时,使用融合了EMA高效多尺度注意力机制的C2f_EMA模块替换主干网络中的C2f模块,提升模型对复杂环境中荷叶病虫害的特征提取能力。试验结果表明,建立的改进YOLOv8荷叶病虫害检测模型能够对荷叶病虫害进行有效检测,实现的平均精度均值(mean average precision,mAP)为89.3%,较基线模型提高了1.6个百分点;模型的参数量较基线模型降低了0.2 M,模型大小仅为5.6 MB。与其他主流检测模型相比,改进YOLOv8模型在检测精度、参数量和模型大小等方面表现出显著优势。将模型部署至Jetson Xavier NX和树莓派4B边缘计算设备上,模型实现的检测帧率分别为27和0.7帧/s,展现了良好的移动端部署前景。所提模型实现了对荷叶病虫害的精准识别,可为荷叶病虫害自动防治提供支撑。

一种轻量级YOLOv5S农作物虫害目标检测模型

[目的]本文提出一种轻量级YOLOv5S农作物虫害目标检测模型以解决在样本数量不足的情况下农作物虫害识别的问题。[方法]利用Ghost技术将2个Ghost Bottle Block线性特征提取模块封装为1个GB模块,代替YOLOv5S中前7个CBL、CSP、SPP非线性特征提取模块,从而约简了YOLOv5S的网络参数,减轻了网络体量。[结果]在保证虫害检测效果的前提下降低网络对计算硬件与训练样本的依赖。为了验证模型的有效性,对水稻、玉米、棉花、马铃薯、苜蓿5种农作物的12类虫害进行识别与定位,多类别平均精度(mAP)为91.31%,比YOLOv5S模型高出2.56百分点。[结论]通过与SSD、Faster-RCNN、YOLOv5S模型对比发现,本文提出的模型在mAP、F1-score、精确率(Precision)、召回率(Recall)4个评价指标均占优势,尤其在小目标、密集目标、与背景相似目标的检测方面表现突出。

计算机视觉技术在农作物病虫害检测中的运用研究

现代农业发展更趋集约化、信息化,这为计算机视觉技术的运用提供了空间。以计算机视觉技术在农作物病虫害检测中的运用优势为切入点,在此基础上分析其具体运用方法,包括建设信息作业系统、强调信息采集、运用可视化技术、重视信息复用等内容。最后就该技术在农作物病虫害检测中的未来运用进行展望,一方面客观呈现技术优势、特点,另一方面为其发挥价值提供必要参考。

基于深度学习的农场虫情检测算法研究及实现

传统的病虫害防治手段需要消耗大量人力、物力,且达不到很高的精确度,为了更科学、高效地做好农场病虫害防治工作,本文结合深度学习技术和物联网技术研发了虫情检测系统进行病虫害的远程检测,提高防治工作的效率。该系统采用YOLO-v5网络模型,结合迁移学习,训练学习了林业常见害虫和农田常见害虫的特征,实现了高效的检测识别。基于物联网技术实现远程控制拍摄病虫害图像,并通过Wi-Fi传输到计算机端进行识别,通过可视化界面呈现出农田中虫害的种类和数量,对减少人力、物力消耗以及实现科学防虫具有良好的实际应用价值。

大田环境下的农业害虫图像小目标检测算法

智能化害虫检测是目标检测技术在农业领域的重要应用,可以有效提高害虫测报工作效率和可靠性,保障农作物产量和质量。在诱虫灯、粘虫板等固定式诱捕装置下,图像背景简单、光照条件稳定、害虫特征显著易于提取,害虫检测可以达到较高的准确率,但其应用场景固定,检测范围局限于设备周围,无法适应复杂的田间环境。针对田间环境下图像背景复杂和害虫尺寸小带来的难检和漏检问题,提出一种改进YOLOv5的小目标害虫检测算法Pest-YOLOv5,以提高害虫测报的灵活性。在特征提取网络中增加坐标注意力机制,通过结合空间和通道信息,增强对小目标害虫特征的提取能力,在颈部连接部分使用双向特征金字塔网络结构,通过融合多尺度特征,缓解多次卷积带来的小目标信息丢失问题。在此基础上,使用SIoU和变焦损失函数计算损失值,同时通过实验得到最优分类损失权重系数,使模型更关注分类困难的目标样本。在公开数据集AgriPest上的实验结果表明,Pest-YOLOv5模型mAP_(0.5)和召回率分别为70.4%和67.8%,优于原YOLOv5s模型、SSD和Faster R-CNN等经典目标检测模型。与YOLOv5s模型相比,Pest-YOLOv5模型将mAP_(0.5)、mAP_(0.50:0.95)和召回率分别提高8.1%、7.9%和12.8%,改善了难检和漏检情况。

基于注意力机制与YOLOv5s的轻量化农作物害虫检测方法

为解决在自然环境中人工检测害虫精度低、速度慢的问题,提出了一种基于注意力机制与YOLOv5s的轻量化目标检测算法.首先,使用Ghost卷积替换YOLOv5s中的普通卷积,得到轻量化的主干特征提取网络.其次,在YOLOv5s中加入加权双向特征融合机制,从而实现高效的双向交叉连接和多尺度特征融合.最后,在主干网络中加入坐标注意力机制,从而增强网络模型对位置信息的关注.与原YOLOv5s算法相比,新算法在IP102农作物害虫检测数据集上的平均精度均值提升了2.1%,模型参数量和计算量分别减少了44.6%和44.3%,检测速度为64.8 FPs.实验结果表明,基于注意力机制与YOLOv5s的轻量化目标检测算法不仅提升了农作物害虫检测精度,而且显著降低了模型参数量和计算量,能够满足农作物害虫检测的应用需求.

基于深度学习的多模态病虫害检测模型训练与验证研究

多模态病虫害数据集具有多样性,导致模型平均精度较低。基于此,文章设计了一种基于深度学习的多模态病虫害检测模型。首先,采集高清图像、音频、视频等多模态数据来整合现有资源构成数据集;其次,进行数据增强以提高模型的泛化能力,减少了过拟合风险;最后,选用深度学习中的MobileNet作为主干网络来捕捉关键特征,模型训练融合了多模态数据,并通过动态权重分配和注意力机制显著增强了检测效能。实验结果表明,在迭代60次时,设计模型的mAP达到0.87,远超文献[1]和文献[2]模型的0.72与0.79,能够快速收敛并准确检测多种病虫害。

基于CNN-Transformer的农作物病虫害知识问答意图识别与槽位填充联合模型

意图识别与槽位填充是农作物病虫害知识问答中问题理解的两个重要任务。在已有面向农业领域的研究中,上述任务仍被视为两个完全独立的子任务,并且未充分利用意图识别与槽位填充的语义信息。为此,该研究提出一种基于CNN-Transformer的意图识别与槽位填充联合模型(CDPCT-IDSF)。该模型根据农作物病虫害文本语义复杂设计CNN网络与多层Transformer结合强调局部的有用信息以缓解语义缺失问题;然后在Transformer解码器中引入对齐保证输入与输出一对一关系以提高识别正确槽位标签的能力。此外,进一步构建了包含20个意图类别、12个槽位类别和11 242条标注样本的农业病虫害知识问答数据集进行对比试验,CDPCT-IDSF模型在该语料库上的槽位填充F1值为94.36%,意图识别精度为92.99%,整体识别精度为87.23%,优于其他对比模型,结果证明了所提模型在农作物病虫害意图识别与槽位填充任务上的有效性,可为面向农作物病虫害的知识问答研究提供理论支撑。

复杂场景下害虫目标检测算法:YOLOv8-Extend

[目的/意义]实现复杂的自然环境下农作物害虫的识别检测,改变当前农业生产过程中依赖于专家人工感官识别判定的现状,提升害虫检测效率和准确率具有重要意义。针对农作物害虫目标检测具有目标小、与农作物拟态、检测准确率低、算法推理速度慢等问题,本研究提出一种基于改进YOLOv8的复杂场景下农作物害虫目标检测算法。[方法]首先通过引入GSConv提高模型的感受野,部分Conv更换为轻量化的幻影卷积(Ghost Convo⁃lution),采用HorBlock捕捉更长期的特征依赖关系,Concat更换为BiFPN(Bi-directional Feature Pyramid Network)更加丰富的特征融合,使用VoVGSCSP模块提升微小目标检测,同时引入CBAM(Convolutional Block Attention Module)注意力机制来强化田间虫害目标特征。然后使用Wise-IoU损失函数更多地关注普通质量样本,提高网络模型的泛化能力和整体性能。之后,对改进后的YOLOv8-Extend模型与YOLOv8原模型、YOLOv5、YOLOv8-GSCONV、YOLOv8-BiFPN、YOLOv8-CBAM进行对比,验证模型检测准确度和精度。最后将模型移植到边缘设备进行推理验证,在实际应用场景中验证模型的有效性。[结果和讨论]YOLOv8-Extend在对比实验中均取得良好的表现,其中与原模型对比实验中,精确率、召回率、mAP@0.5和mAP@0.5∶0.95评价指标分别提升2.6%、3.6%、2.4%和7.2%,表现突出,具有更好的检测效果。改进前后的模型分别运行在边缘计算设备JETSON ORIN NX 16 GB上并通过TensorRT加速相比,mAP@0.5提升4.6%,达到57.6 FPS,满足实时性检测要求。在复杂农业场景中YOLOv8-Extend模型具有更好的适应性,在实际采集数据中微小害虫与生长环境相似的害虫检测方面有明显优势,在困难数据检测方面准确率提高了11.9%。[结论]本研究提出的YOLOv8改进模型有效提高了检测精度和识别率同时保持了较高的运行效率,能够部署在边缘终端计算设备上实现农作物害虫的实时检测,也为其他小目标智能检测和模型结构优化提供参考和帮助。

基于改进YOLOv5s模型的柑橘病虫害识别方法

针对现有检测模型不能满足在自然环境中准确识别多种类柑橘病虫害的问题,提出一种基于改进YOLOv5s模型的常见柑橘病虫害检测方法。改进模型引入ConvNeXtV2模型,构建一个CXV2模块替换YOLOv5s的C3模块,增强提取特征的多样性;添加了动态检测头DYHEAD,提高模型对不同空间尺度、不同任务目标的处理能力;采用CARAFE上采样模块,提高特征提取效率。结果显示,改进后的YOLOv5s-CDC的召回率和平均精度均值分别为81.6%、87.3%,比原模型分别提高了4.9、3.4百分点。与其他YOLO系列模型在多个场景下的检测对比,具有更高的准确率和较强的鲁棒性。结果表明,该方法可用于自然复杂环境下的柑橘病虫害的检测。

基于改进Mask-RCNN算法的作物害虫分类识别

智能虫情测报灯对农业生产中及时察觉虫害、虫灾问题有重大作用,准确的害虫分类识别为虫情测报提供可靠数据支撑的关键。该研究对智能虫情测报灯所需核心识别算法进行改进,针对分类目标多尺度、存在多种相似非目标害虫干扰、易产生目标粘连等问题,提出一种基于改进Mask-RCNN(mask region-based convolutional neural network)模型的害虫图像智能识别模型。该模型使用DeAnchor算法改进Mask-RCNN的锚框引导机制,使用NDCC(novelty detection consistent classifiers)训练分类器进行联合分类和检测,改善非目标杂虫的误识别问题。改进后模型对无杂虫、不同虫体密度图像的识别准确率最高达到96.1%,最密集时可达90.6%,在仅有非目标的图片识别中,误检率降至9%,非目标与目标共存且密度为40虫/图的误检率降至15%。试验表明,该文所提模型在现有分类模型的基础上,增强了对密集区域的检测能力,改善了非目标误识别问题,在实际检测环境下的害虫分类识别精度更高,可为虫害防治工作提供数据参考。

基于改进YOLOv5s的自然场景下生姜叶片病虫害识别

为提高自然环境下生姜叶片病虫害的识别精确率,提出一种基于改进YOLOv5s的生姜叶片病虫害识别模型。建立了田间不同自然环境条件下的生姜叶片病虫害数据集,为保证模型在田间移动设备上流畅运行,实现网络模型的轻量化,在YOLOv5s中引入GhostNet网络中的Ghost模块和Ghost BottleNeck结构。同时,为避免生姜叶片病虫害图像小目标特征丢失的情况,增强图像特征提取,加入CA注意力机制模块,提升生姜叶片病虫害的识别准确率和定位精确度。改进后的模型参数量、计算量和权重文件大小分别为YOLOv5s模型的52.0%、50.6%和55.2%,对生姜叶片病虫害识别平均精度均值达到了83.8%。与Faster-RCNN、SSD、YOLOv4、YOLOv5s、Tea-YOLOv5s等算法相比,平均精度均值分别提高37.6、39.1、22.5、1.5、0.7个百分点,将改进后的目标检测模型部署在Jetson Orin NX开发板上,并使用TensorRT、Int8量化和CUDA等方法对检测模型加速,加速后的模型检测速度为74.3帧/s,满足实时检测的要求,测试结果显示,改进后的模型减少了漏检、误检的情况,并且对目标定位更加精准,适用于自然环境下生姜叶片病虫害的精准识别,为后续生姜机械自动化施药作业提供技术理论支持。

基于改进的U-Net网络模型的叶片病害检测

为了满足作物病虫害绿色防治对病虫害程度检测的需求,设计了一种改进的U-Net网络模型用于作物叶片病虫害程度的检测。首先,选择ResNet50网络作为模型的主干网络,借助迁移学习来提升训练收敛速度和降低计算成本。其次,引入注意力机制对U-Net网络的各层特征提取和融合进行优化,以提高网络模型接收关键信息的能力。实验结果表明,改进的U-Net512网络模型具有最优的检测性能,平均检测精度达到90.14%,平均绝对误差为276.3。通过分析模型不同采样深度下的各层特征图发现,注意力机制的引入使网络模型能够获取并融合叶片整体特征和病害区域特征两个维度的信息,进一步提升模型检测性能。这种方法不仅能够有效地检测作物叶片的病虫害程度,而且具有较高的准确性和可靠性,有助于实现作物病虫害的绿色防治。