Study On The Pest And Disease Recognition Model Of Vegetable Leaves Based On Feature Detection

Vegetables are one of the world's most important food crops.Due to the worsening global climate,the economic losses caused by pests and diseases of vegetables continue to increase every year.The specific implementation and application of image processing technology based on feature detection in the field of leaf disease and insect pest recognition is of great significance to the development of information agriculture,and it is also the research focus and focus of smart agriculture and agricultural expert systems.This paper addresses the problems of image leaf pests and vegetables in the process of image collection,which are susceptible to noise and cause obvious loss of picture details.The traditional feature extraction algorithm has low initial feature point selection.The county's two bases in Liushui Township took cucumber,Chinese cabbage and potato as experimental objects,and collected 1,000 vegetable leaf disease images and 500 pest image images.The main research contents and results are as follows:(1)The image filtering algorithm is studied.According to the characteristics of the experimental objects and three kinds of filtering algorithms,the gaussian filtering algorithm is selected to carry out the image denoising work in this paper.The data of two bases in Xiyan Village,Misha Town,Dehui City,Jilin Province,and Liushui Township,Changling County,Songyuan City,Jilin Province were collected and processed.The collected images of vegetable leaf diseases and insect pests were cropped to the same size,and similar images were classified class.Aiming at the defect of randomness of standard deviation artificially during image denoising using traditional Gaussian filtering algorithm,an improved adaptive Gaussian filtering algorithm was researched.Using cucumber,Chinese cabbage and potato leaf disease images as experimental objects,comparative experiments of traditional algorithm,reference 34 algorithm and improved algorithm were carried out.The experimental results show that the PSNR value of the improved algorithm is 13.5% higher on average than the PSNR value of the literature 34 algorithm.(2)The feature extraction algorithm of image texture and color is studied.The Canny algorithm,SURF algorithm,and the HSV color histogram were used to extract the texture and color features of three vegetable leaf disease and pest images.Aiming at the problem that the accuracy of the initial feature point selection of the traditional SURF algorithm is insufficient,an improved SURF feature detection algorithm is proposed.Based on the images of common diseases and insect pests of cucumber,Chinese cabbage and potato,experimental experiments were performed to compare the improved SURF algorithm with the traditional SURF algorithm.The experimental results show that the repetition rate of the improved SURF algorithm for detecting image characteristics of three vegetable leaf diseases and insect pests is above 85%,which is higher than the traditional SURF algorithm.(3)A vegetable leaf pest identification model based on feature detection was constructed.The SURF algorithm and Canny algorithm were used to extract the texture features of vegetable leaf diseases and insect pests.The HSV color space was used to extract the color features of vegetable leaf diseases and insect pests.The texture feature vector and color feature vector were sorted according to the ratio of 3: 1,30 feature symptoms were extracted,and 15 disease types were summarized,and a feature vector library of images was constructed.Using the characteristics of three vegetable leaf disease and pest image as input parameters of the network model,a vegetable leaf disease and pest recognition model was constructed,which was divided into two sub-networks of disease and pest respectively for training,and the trained model was used for test experiments.The experimental results show that: the average accuracy rate of identification of 10 types of leaf disease types in cucumber reached 93.42%,the average accuracy rate of identification of 5 types of leaf disease types reached 93.44%;the average accuracy rate of identification of 10 types of leaf disease types in Chinese cabbage It reached 91.46%,and the average accuracy rate of the identification of 5 types of leaf pest types reached 91.32%;the average accuracy of the identification of 10 types of potato leaf disease types reached 91.42%,and the average accuracy of the identification of 5 types of leaf pest types reached 92.20 %.A comparative analysis between the traditional model and the improved model was conducted,and the results showed that the improved model improved the average accuracy of identifying pests and diseases of three vegetable leaves by an average of 5.48% on the premise of slightly discarding some operating efficiency.