The Recognition System Of Cotton Pests Based On Digita Mage Processing Technologies

Cotton is an important economic crop in China, it underwent the most rapidly growing period in the twentieth century, and even the total production topped the world in the twenty-first century. Accordingly, cotton technologies also made new breakthrough and reached the advanced level in many aspects around the world. However, cotton pests control haven’t made great progress for a long time, what is worse, the perennial loss caused by cotton pests takes up 15 to 20 percent of the total production. It still relies on naked eye to identify the cotton pests, though there are a great many species of pests in the cotton field so far. Such backward method of identification is the main cause of the slow progress of the pest control technology. Thus it is becoming more and more important to classify the pests in a scientific and reasonable way and with an accurate and efficient technique. As computer technologies develop so rapidly, so does the digital image processing technologies, which depends on the computer technologies to simulate the visual system of human being by image de-nosing, segmentation, feature extraction as well as classification analysis so as to achieve the aim of understanding and identifying the objects. Therefore the application of digital image process techniques into the automatic identification and classification of the cotton pests plays a significant role in the targeted pest control, avoidance of economic loss as well as maintenance of ecological balance.This paper mainly discusses the pest classification methods with the application of various image digital process techniques into pest species which differentiated by their external morphological characteristics. The test firstly samples fifteen kinds of imagoes in the upper-middle reaches of Yangtze River cotton filed and then applies the grayscale conversion method according to the morphological feature of the pests, chooses the best conversion of image details based on the conversion results, then denoising and enhance the above converted image by median filtering and histogram equalization processing, and segment the pretreated image of left and right wings by using Canny algorithm and improved mathematical morphology method; Then segment the adversary, chest, abdomen, foot and tentacles by means of Otsu threshold segmentation method, extract the mathematical morphological parameters of the geometry for the segmented images by using a mathematical algorithm, and then achieve the identification of cotton pest species according to the mathematical parameter design classifier.The main results are as follows:1. Pest image preprocessing:the effect of conversion using blue components as grey value in the process of image preprocessing is the best and the image detail is clearer than other methods; the converted image denoising and enhancement with median filtering and histogram equalization can significantly increase the image display details and quality.2. The image and integration:compared with Sobel operator and the Log operator, Canny operator testing works best in pest wings segmentation with more completed, less burr and less noise; The use of the improved algorithm of mathematical morphology and feedback can obviously improve detail display and clarity of pest wings image after the segmentation.3. The extraction of feature parameters:by the use of the basic mathematical algorithms of geometry, the external morphological characteristic parameters can be successfully extracted with high reliability and real and reasonable reflection after many-time extractions and data comparisons.4. Classifier design:Cole Mo Rove test is adopted to estimate whether it accords with normal distribution, and then it will be discussed whether classification method is feasible according to the estimated result; Fuzzy clustering algorithm is used for the cotton pests which accord with basic normal distribution, while binary tree and vector machine (SVM) classification method is feasible for the pests which donot accord with the normal distribution, then classification results can be obtained.The above results show that all kinds of digital image processing technologies and methods used in the tests have a good application effect in the process of cotton pests’ identification, therefore can dramatically improve the efficiency and accuracy of the pests’pictureidentification.