Research On Locust Recognition Algorithm Based On Image Processing Technology

With the development of machine learning and deep learning,image recognition and target detection are widely concerned.Locust detection and recognition,as the basis of studying locust outbreak mechanism and establishing prediction model,is an important means of establishing dynamic monitoring and early warning system of locust disaster.At present,there are few researches in the field of locust recognition and they are relatively lagging behind,and differences are obvious in recognition accuracy and speed between traditional recognition methods and new target detection methods.In addition,the traditional research methods are mostly suitable for the detection of locusts in a single background,and the recognition accuracy of locusts in natural background such as grassland and crops is low.Based on this problem,the region significance recognition algorithm based on binary image,segmentation algorithm by fusion of multiple colors and faster RCNN algorithm based on candidate region were used to research on detecting and recognizing of locust image for the purpose of improving the accuracy of locust recognition in complex background,and achieved better results.This paper presents region significance recognition algorithm based on binary image,which extracts local features of locust and background to realize binary classification by support vector machine.Experiments show that the algorithm can accurately recognize locusts in the same area.In order to segment a large area ofadherent locusts,an algorithm by fusion of multiple colors is proposed.The algorithm classifies based on each pixel,instead of focusing on the individual characteristics of locust.Considering that locusts and backgrounds are in different color ranges,so we convert the image to a different color model for fully enlarging their color difference,and finally completed the image segmentation based on pixels.In order to recognize locusts in dry period,soil,rock and other environments.Faster RCNN algorithm can also achieve accurate recognition in the case that locust and environment color are very similar,which complements the limitations of the two algorithms.In this paper,the detection of locust images in different periods,different environments and different touching degree are completed,which enriches the identification methods,improves the recognition accuracy and speed effectively,and meets the requirements of detection warning system.