Classification And Identification Of Corn Ear Based On Deep Learning

With the rapid development of modern information technology,computer vision has achieved considerable results in different fields.Image recognition,as a key link in the computer vision technology system,is characterized by the need to rely on open data for more data,open source for more basic tools,and update iterations of algorithms.The use of image recognition and computer simulation technology to simulate human classification and judgment of observed crop images has become an indispensable step in smart agricultural applications.The image recognition of crop ears represented by corn can realize intelligent operation such as precise crop cultivation and automated seeding,which greatly reduces labor cost and improves classification recognition accuracy.Among them,deep learning,as a new subject in the field of machine learning,its image application based on convolutional neural networks has achieved remarkable results in the field of image recognition.Therefore,in this paper,the deep learning application is incorporated into the image recognition of corn ears,and the corn grain and cob images are classified and identified.The specific research contents are as follows:1.Introduce the current status and problems of deep learning development.Through the Web of knowledge,the relevant literature data on the identification of maize varieties in the past 8 years was used to analyze the research on the identification of maize varieties at home and abroad.2.Set up the image acquisition device to build the data set.Grain and cob samples were obtained by artificially testing corn ears in an indoor environment.In order to meet batch processing and post-experimental operations,this experiment used the corn kernel phenotypic detection system developed by Beijing Agricultural Information Technology Research Center to image pre-process the acquired images to obtain experimental image datasets containing target features.3.In-depth study on the classification algorithm of corn grain image based on deep learning.The corn kernel integrity identification was carried out,and pictures of different sizes were constructed to compare the influence of image size difference on the recognition rate.Further improve the network performance and expand the network.The corn grain image recognition results based on convolutional neural network are compared with the results of traditional corn grain image recognition results,and the appropriate activation functions and methods are selected.The experimental results verify that the grain dataset is94.6% under the VGG16 deep network,which is significantly higher than the traditional neural network.Practice has proved that this method has good robustness and high recognition rate.4.The Mask R-CNN segmentation algorithm was used to detect the corn cob cross-sectional image through the target detection framework.The design method ofconstructing corn cob core detection based on Mask R-CNN regional convolutional neural network was introduced in detail.ResNet101 was used as the classification method.The model is trained on the corn cob image dataset.The experimental results show that the mAP value of the method reaches 0.94,which can achieve efficient and accurate regional detection on multiple datasets.