Research On Intelligent Fruit Picking Recognition System Based On Hi3559AV100

As a direction of the development of smart agriculture,the emergence of picking robots undoubtedly provides solutions to the problems of insufficient labor,expensive picking operations,large picking workloads,and low picking efficiency faced in agricultural production.It also provides inexhaustible impetus for the sustainable development of agriculture.As the key technology of intelligent fruit picking,automatic fruit recognition determines the performance of the picking system.Based on the theory of deep learning,this paper uses the YOLOv3 deep learning model to extract features of the shape and color of various fruits to realize automatic fruit recognition.In order to achieve the purpose of real-time application,this subject has realized a fruit automatic identification system based on Hi Silicon Hi3559AV100 embedded platform.First,by comparing and analyzing the advantages and disadvantages of several mainstream target detection algorithms,the YOLOv3 algorithm is used as the main body of the fruit detection and recognition part,and automatic fruit recognition is realized on the PC side.The simulation results show that the average recognition rate of the three fruits is 98% Above;followed by model conversion and accuracy testing.Convert the YOLOv3 algorithm model implemented on the PC side into a Caffe-YOLOv3 model,and compare and analyze the model detection accuracy before and after the conversion to ensure that there is no accuracy loss;then in the Linux operating system Set up the Hi Silicon development environment,use the NNIE hardware resources in Hi3559,combined with the fast performance of YOLOv3,to realize the automatic identification and positioning of the three kinds of fruits.The test results on the Hi3559 embedded hardware platform show that the average recognition accuracy rate is over 80%,and the recognition speed reaches 5-8fps,achieving practical purposes.