Research On Recognition System Of Rice Field Pests Based On Machine Learning

The outbreak of pests often leads to the reduction or even stagnation of crop yield.Therefore,the recognition and control of pests has always been the most important task in China’s agricultural basic work.The rapid identification of pests can effectively prevent and control the outbreak of pests and reduce food loss,which is of great practical significance.At present,traditional pest recognition work is mainly carried out by experienced professionals,and then prevention and guidance work.However,the outbreak range of pests is generally large and the propagation speed is fast.The traditional pest recognition method is inefficient and it can’t complete the prevention work well.In recent years,artificial intelligence technology has ushered in the peak of development.Machine learning,as one of the important realization methods in the field of artificial intelligence,has also been widely applied and studied.More and more scholars have conducted in-depth research on machine learning in the field of crop pest recognition,it is expected to provide guidance to farmers in time,reduce the economic losses caused by pests,and provide theoretical basis for remote diagnosis and prevention of pests in precision agriculture and crops.In this paper,the operator of Haar-like feature description is used to extract the features of pest images of rice field.The extracted features are used to build a weak classifier,and Ada Boost algorithm is used to build a set of weak classifiers to get a strong classifier.Finally,an Ada Boost cascade classifier is used to identify pests.The research content of this paper is mainly based on the following aspects:(1)According to the characteristics of small size and small area of pests in the picture,in this paper,a set of flow is designed which combines image filtering,image binarization,edge detection and other image processing methods with image segmentation.Finally,the foreground region including rice pests was segmented from rice pest image,and most of the background regions were removed,which improved the recognition efficiency.(2)In this paper,the operator of Haar-like feature description is studied theoretically.The feature matrix of Haar-like is used to extract the rectangular feature in the pest image,and then the Ada Boost algorithm is studied theoretically.Based on the extracted rectangular feature,the Ada Boost weak classifier is trained,then the weak classifier set is constructed to get the strong classifier,and the strong classifier is cascaded to obtain an Ada Boost cascaded classifier.Finally,the structure of the cascade classifier is analyzed and studied,and the structure of the cascade classifier is improved.In the experiment,four groups with different experimental conditions were set up to identify five rice pests.The results show that the average correct recognition rate of the improved cascade classifier based on image segmentation is 95%,and the average error recognition rate is 7.94%,which is the best experimental result in all the grouping experiments.(3)The rice field pest recognition system is designed and completed by choosing the mainstream microservice architecture as the system architecture and taking Spring Boot and Spring Cloud as the development framework.The system can not only provide the recognition function of rice pests,but also include the introduction of rice pests,pest damage articles,pest control and other functions.