苹果树病虫害智能识别系统设计与实现

为提高苹果的产量和质量,防止病虫害对果实质量的影响,设计了一款基于机器视觉的苹果树病虫害智能识别系统。该系统采用交互式分割(GrabCut)算法对图像进行分割,然后使用高斯拉普拉斯算子和拉普拉斯高斯(Laplacian-of-Gaussian, LOG)算法将苹果叶片中的病斑提取出来,最后将提取出的图像送入深度神经网络(deep neural networks, DNN)进行进一步的分析与处理,能够实时、方便地识别出苹果树叶病害中较为常见、发病率高的花叶病,锈病,灰斑病,斑点落叶病以及褐斑病。经测试,该系统对苹果树5种常见病虫害识别率精度高达91.17%。结果表明,该算法能够有效提升苹果树病虫害防治,优于基于卷积神经网络特征的区域方法(regions with CNN features, R-CNN)、YOLO(you only look once)等单一病虫害检测方法。

Design and Implementation of Intelligent Recognition System for Apple Tree Diseases and Pests

In order to improve the yield and quality of apples and prevent the influence of pests and diseases on fruit quality, this paper designs an intelligent identification system for apple tree pests and diseases based on machine vision.The system uses the GrabCut algorithm to segment the image, then uses the Laplacian of Gaussian operator and the LOG algorithm to extract the lesions in the apple leaf, and finally sends the extracted image to the DNN neural network for further analysis and processing. It can identify the common and high incidence of mosaic disease, rust, gray spot, spotted leaf disease and brown spot in apple leaf diseases in real time and conveniently. After testing, the system has an accuracy of 91.17% for the recognition rate of 5 common diseases and pests of apple trees. The results show that the algorithm can effectively improve the control of apple tree diseases and insect pests, and is better than RCNN, YOLO and other single pest detection methods.