Research On Crop Counting Method Based On Convolutional Neural Network

Compared with traditional crop counting methods,image processing technology is used to extract color and texture features,and machine learning method is used to est ablish regression model.The method based on deep convolution neural network has th e advantages of simple process,good recognition effect and simple subsequent countin g.From the point of view of in-depth learning,after analyzing and summarizing the e xisting crop counting methods,this paper realizes the crop counting method based on color characteristics for different crop growth cycles in the field environment,and puts forward the crop counting method based on improved YOLO?V3 and the crop counti ng method based on multi-column convolution neural network.Overall,the main work of this paper is as follows:(1)After consulting a large number of literatures,the traditional crop counting me thods are summarized,and the main methods are reproduced.In addition,this paper a lso summarizes the methods of crop counting based on convolution neural network,an d elaborates on the main methods.(2)By using image processing technology,the yield of the counting method based on color features is proposed.Taking the wheat ear image as an example,and use thi s method to a series of digital image processing technology the processing section and the ear is H picture,then use the fast parallel thinning algorithm to improve the ima ge and extract the skeleton,and the Harris corner detection and corner detection.The screening results meet the requirements.The practical application.(3)This paper improves YOLO?V3 network,which is one of the most commonly used networks in the field of target detection and recognition.Dense Net is integratedinto YOLO?V3 network,and YOLO?D network is proposed to make the network rec eive more image features.The experimental results on the self-built data set of rice se edlings show that the algorithm is effective.(4)In view of the heading stage of crop growth cycle,the field environment is c omplex,the posture of ear organs is diverse,the overlap between ears and spikes is more frequent,and the color overlap is more common.This paper presents a crop cou nting method based on multi-column convolution neural network.The design of this m ethod refers to the idea of end-to-end,and uses local regression network and truth de nsity graph.The network can count directly through integral output.A maize tassel da ta set was constructed using UAV images,and the algorithm was tested on the data s et,and some results were obtained.