基于SVM和区域生长结合算法的南方主要蔬菜害虫分类识别

该文基于支持向量机（support vector machine，SVM）与区域生长结合算法，设计了对黄曲条跳甲、烟粉虱、小菜蛾、蓟马这四类蔬菜害虫进行分类识别的检测算法。该方案将识别过程融入到分割中，采用网格法进行区域生长种子点的选取，简化图像处理的步骤。该文每种蔬菜害虫训练样本图像为60幅，测试样本为40幅。试验展示，基于其形态、颜色特征，该算法可以将南方重大蔬菜害虫正确分割识别出来，对黄曲条跳甲、烟粉虱、小菜蛾、蓟马成功率为分别为96.4%、93.2%、95.4%、98.2%，算法达到了对多种害虫进行分类的效果，有较好的应用前景。

Classification and recognition for major vegetable pests in Southern China using SVM and region growing algorithm

Abstract: As vegetable safety is a rather important issue related to people''s health and life, it is fundamental to ensure the vegetable safety by supervising the whole procedure of vegetable production. This requires to control the use of pesticides via accurate pesticide spraying according to the pest situation, which is the best strategy for vegetable safety. The key issue achieving this objective is to find out the species, quantity, and distribution of pests and the harm degree of vegetables. Although the pest identification via image processing has been widely used in recent years, it merely handles small pests of vegetables in the laboratory, and the number of pest species simultaneously processed is also limited to 1 or 2. To better recognize pests, this paper proposes a new algorithm to identify a number of vegetable pests such as striped flea beetle, whiteflies, diamondback moth, and thrips by deploying the support vector machine (SVM) and the region growing algorithm. This scheme integrates the recognition process into the segmentation one and uses the grid method to select seed points for region growing, which in turn simplifies stages of image processing. In performance assessment, 100 samples are adopted for each test pest, among which 60 are for training and the others for testing. Experimental results show that the proposed scheme can correctly identify the aforementioned major 4 vegetable pests in south China with a recognition rate of more than 93%. This implies that the proposed scheme achieves classification of several pests and thus would be promising in practical applications.