基于图像分割和轮廓矩的果蝇求偶行为识别方法

摘要

监控害虫求偶行为是农业病虫害防治的重要手段，传统的人工识别工作量大、效率低下，为实现自动检测与识别果蝇求偶行为，设计了基于图像分割和轮廓矩的识别方法。该方法首先采用基于灰度图像的背景生成、更新和阈值分割的方法自适应地获取果蝇的轮廓。然后根据轮廓矩理论，提取各果蝇轮廓的矩不变量并输入BP神经网络判断是否发生单侧振翅行为。最后根据果蝇间的位置关系，判断振翅果蝇是否在尾随其他果蝇，从而确定其是否发生了求偶行为。采用人工检测的方法进行对比验证，结果表明该方法能够有效地识别果蝇的求偶行为，准确识别率达89．6％。%Monitoring pests’courtship behaviours is a major means in agricultural pest and diseases management.Traditional manual identification has heavy workload and low efficiency.In order to detect and identify the courtship behaviours of drosophilae automatically,we design an identification method which is based on image segmentation and contour moment.First,the method employs the grayscale image-based methods of background generation,update and threshold segmentation to adaptively obtain the contours of drosophila.Then according to the theory of contour moments,it extracts the contour moment invariants of every drosophila contour and inputs them into BP neural network to estimate whether the drosophila vibrates its unilateral wing.In third step,it judges whether or not the wing-vibrating drosophila is trailing other drosophilae according to the position relationship between them so as to determine if drosophila has the courtship behaviour happened. Contrast verification is carried out in manual detection way,results show that the new method can effectively identify the courtship behaviour of drosophila and the accuracy of recognition reaches 89.6%.