Study On Identification Method Of Agricultural Pests Based On Machine Vision

Pest monitoring and pest statistical forecasting is an important part of integrated pest management prevention traditional pest monitoring and statistical methods need the people who has professionals knowledge to complete, and it will spend a lot of human and financial resources. The machine vision-based pest identification is an efficient, non-contact and high recognition accuracy methods, it is suitable for the modern agricultural production.Pest identification and counting techniques based on machine vision applied entomology, image processing and pattern recognition. The current study is aimed primarily pest identification in laboratory environment, and the laboratory environment is more stable. In the laboratory, there is no wind and debris, the light intensity is uniform and unchanged, and the pest samples are usually placed by hand. Therefore, in this condition, the background for the pest image captured is simple, light is stability; the target pest posture is good. In order to achieve high accuracy and efficient with the machine vision-based classification and identification of pests in the complexity and actual agriculture environment, to improve and perfect the problems and shortcomings.Pest identification technology included:image acquisition, image preprocessing, feature extraction, feature optimization, classification and several other steps. In order to obtain a more accurate and reliable characterization data for improved image preprocessing techniques to solve the problem of pest placing consistency, reducing errors of gray texture feature data. Image feature is the basis for the identification of pests, in this paper, has extracted morphological features, and build 54 dimensions space included 17 kinds of original feature,26 kinds of texture features and 11 kinds of color features. In order to make the pest classification and identification results more accurate and high efficient, using a variety of intelligent algorithms to optimize and reduce the original feature space dimensionality, extracted optimal feature subspace to improve the performance of pest identification.In order to verify the feasibility of the above design methods, we use industrial machines, cameras, lighting, and metal chassis and other hardware devices developed for field environments and automatic identification of pests counting system. System software using C# language, with the hardware to achieve the automatic identification and counting of pests, pest model, to establish real-time, remote transmission of information on pests and other functions, and preliminary tests.