Crop Pest Image Classification Based On Significance Detection And BOW Model

Crop pests are various and prone to outbreaks,which not only affect farmers’ economic benefits,but also have a certain impact on social stability.Therefore,how to effectively identify crop pests is of great importance.At present,there have been a large number of studies on crop pests at home and abroad,but most of them are targeted at a certain kind of pest,which is not applicable.However,due to the variety of crop pests and the lack of a complete statistics,it is difficult to identify and detect crop pests.With the development of computer vision technology and artificial intelligence,scholars at home and abroad have conducted a lot of research on the combination of computer vision technology and pest classification and identification,which is also the hot and difficult point of expert research.Bag-Of-Words model was first applied in the field of text research.In recent years,the combination of word bag model and computer vision technology applied to the research of image classification has proved the superiority of BOW model in the field of image.In this context,combining significance detection and BOW model,this paper took 2200 pest images collected as experimental samples to explore and study the classification and identification of crop pests.The main results of this paper are as follows:(1)The significance detection algorithm is studied in this paper.The six significance detection algorithms AC,FT,HC,LC,ITTI and GBVS were compared and analyzed.Based on the principle of preserving the integrity of pest targets,the advantages and disadvantages of the six algorithms were compared.Finally,GBVS algorithm was selected to calculate the significance map of crop pest images.(2)This paper realizes the region of interest extraction.Firstly,threshold segmentation is carried out on the significant graph obtained by using the method of maximum inter-class variance to obtain the binary graph,and then the region of interest of the image is obtained by combining the binary graph with the original image.The experimental results show that the interference of background to the target classification can be effectively reduced by extracting features from the region of interest rather than directly from the image.(3)This paper realizes the modeling of Bag-Of-Words model and the training of SVM classifier.Surf algorithm was used to extract the features of the regions of interest.K-means algorithm was used to cluster the extracted features.Finally,SVM support vector machine was used to complete the training and 22 binary classifiers were obtained.The experimental results show that the recognition accuracy of this method is nearly 10%higher than that of the traditional word bag model method.(4)The effects of parameters in crop pest classification and identification on training results are analyzed,and the optimal parameters are selected.Through continuous testing,when K value was 1200,the classification effect was the best,reaching 60.27%,among which the rice leaf roller had the highest classification accuracy,reaching 97%.Whitefly was next,the classification accuracy was 92%.Secondly,the effects of different feature extraction algorithms on classification recognition were discussed.The experimental results showed that surf feature-based Bag-Of-Words model had a higher accuracy than sift feature-based Bag-Of-Words model in classification recognition.Through analysis and calculation,it was found that surf feature extraction would be able to better express the target,regardless of the method in this paper or the traditional word bag model.