摘 要

农业与民生和经济发展息息相关，对农业发展科学化的关注既是民生需求， 也是经济稳步发展的迫切需求。病虫害是影响农作物生长的重要因素，对农作物的产量和品质都能造成无法估计的损害。

针对目前广大农业产区农业植保人员稀缺、病虫害识难度大等问题，论文在当前已有的研究基础上进行设计，提出了一套基于机器视觉的远程害虫种类识别和数量检测系统，该系统能够在农业产区进行害虫捕杀和图像采集，同时将昆虫图像上传到虫类鉴别服务器进行昆虫识别。论文以机器视觉为核心，利用 OpenCV 开源视觉库，研究和实现了昆虫识别的完整流程。论文研究的内容主要包括：

（1） 对昆虫图像进行图像预处理的研究。使用了加权平均法对图像进行灰度化处理，再使用高斯滤波对图像进行平滑处理，最后使用大津法对灰度化后的图像进行二值化处理。

（2） 对昆虫图像进行特征提取的研究。使用 OpenCV 对二值化后的图像进行轮廓查找，同时实现昆虫计数，然后分别针对昆虫轮廓的矩形度、延长度、似圆度、球状性、叶状性进行数学定义和特征提取。

（3） 对昆虫识别进行了分类器的研究。选取了逻辑斯蒂回归模型、线性 SVM 模型和 K 临近模型进行分类器的训练和测试，比较了三种分类器在昆虫识别上的性能。

论文对基于机器视觉的昆虫种类及数量检测机制进行了研究，并在此基础实现了昆虫的识别和计数。

关键词：昆虫识别；机器视觉；图像处理；机器学习；

ABSTRACT

Agriculture is closely related to people’s livelihood and economic development. Scientific research on the development of agriculture is not only the need for people’s livelihood, but also the urgent need for steady economic development. Pests are important factors that affecting the growth of crops, and pests can cause unpredictable damage to the yield and quality of crops.

To deal with the problems such as the scarcity of agricultural plant protection per- sonnel and the difficulty of identification for pest and disease, This paper presents a set of remote pest species identification and quantity detection system based on machine vision, which is based on the existing research. This system is capable of pest killing and image acquisition in agriculture areas, while uploading pest images to insect iden- tification servers for insect identification. This paper the use of OpenCV open source vision library. This paper includes the following contents:

(1) Research on image preprocessing of insect image. Use the weighted average method to gray the image, and then use Gaussian filtering to smooth the image. Finally, binarize the gray scale image by OTSU method.

(2) Research on Feature Extraction of insect image. Use OpenCV to find the contour of binarized image, while realized the insect counting. Then propose the mathematical definition of insect features and extract the insect features, insect features include rec- tangularity, elongation, roundness, sphericity, leafy.

(3) Research on Classifier for insect identification. This paper chose the logistic regres- sion model, the linear SVM model and the k-nearest neighbors model as the insect clas- sifier to training and testing. Then compare the performance of three classifiers on in- sect identification.

This paper studies the counting and identification of insects based on machine vi- sion, achieve the counting and identification of insects.

Key words : Insects Identification; Machine Vision; Image Processing; Machine

Learning

目 录