The Research Of Recognition And Detection Of Oxya Chinensis Larva Based On Computer Vison

Paddy is a major food crop in China.With the increasing requirement of the quality and yield by people,more refined management of the crop is needed to decrease the pesticide use.Oxya chinensis is dominant species in paddy field's locust,almost 80%of the total locust.Traditionally,the pest census by experts,which is time-consuming and labor exhaustive.Basis of Oxya chinensis habits of highly gregarious at the early age.In this paper,we concentrated on Oxya chinensis trapped by sticky plank.Then,we achieved the image of the plank,even more the image in natural environment.Image processing,pattern recognition,target detection technology of computer were used in the paper to research and identification the pests.In this paper,we distinguish the age of the early third age of the Oxya chinensis,and recognition the major pest in the paddy field on the trapped plank,including Cnaphalocrocis medinalis,Chilo suppressalis and the early age Oxya chinensis.Furthermore,we train a model by DPM to detect the Oxya chinensis in the natural environment and on the trapped plank for a more accurate research in the future.In this thesis,the research centred on the Typical paddy pests-Oxya chinensis.We studied the technologies of image segmentation,feature extraction,pattern recognition,classification and detection based on pests' images.The main research contents of thesis include:(1)We designed a image acquisition system aim at trap the Oxya chinensis and the overwintering pest in paddy field to get the pest image.The distance of plank from the camera,and the scope of vison were fixed.In this research,the pests on the plank always have strong activity and the posture of the same species changed rapidly,the length of different instar also is great different.The segmentation technology of the object pest on the trapped plank image based on HSV color model's V channel to enhance the original image to a better segmentation result was put forward.According to the characteristics of the objectives and the plank environment,enhanced the image before the Otsu threshold segmentation was applied.Before the Otsu algorithm segmentation,we extracted the HSV model's V channel to enhanced the original image to enhance gray level between the objective pests and the background.Then used the Otsu algorithm to get the threshold adaptively of the enhanced pest image.After finishing the segmentation,morphological processing was needed to get the whole pests for the subsequent processing.Then,based shape of the growth cycle of Oxya chinensis,we distinguish the age of the pest for the online detection and management used in the crop field.(2)The pattern recognition was studied.The SVM model was used to identify the pests achieved form the trapped plank.We introduced the basic methods of support vector machine(SVM),which we used to classification the early pests in paddy field.The pests' image normalized in same size and segment by extracted the biggest connected domain.Multi-feature was extracted after segmentation,what includs morphological,color and texture characteristics.The identification experiments achieved the accuracy was 88.3%.(3)Same to human detection,pest detection itself has flexible variability and the environment is complex,so these object detection is a challenge topic.So far,there are many human detection methods.Deformable part models(DPM)is one of the best object detection algorithms according to VOC challenges.In this paper,we proposed using DPM to train a model to detect the objective pest in natural environment and on the trapped plank,to offer a theory support for the future more accurate research.