Robot Flower Sorting System Based On Deep Separation Convolutional Neural Network

The application of low-latency and high-precision image classification algorithms in industrial robot-assisted embedded devices has always been one of the key technical issues and scientific research hotspots in practical engineering.At present,the more mature image classification algorithms are based on traditional image processing technology and image classification technology based on deep learning.However,the transplantation image classification algorithm on micro-embedded devices has certain consistency in maintaining high image classification accuracy and real-time performance.limits.Therefore,it is of great research significance to study the image classification technology based on embedded devices from the new perspective of embedded device hardware and software design and optimization of lightweight convolutional neural network model.In this thesis,the industrial flower automatic sorting operation is taken as the engineering background,and the micro-embedded equipment is taken as the specific research object.According to the requirements of this system,the idea of optimizing the lightweight network model is introduced.Based on the actual project requirements of micro-embedded devices,the selection of image classification network model,and model optimization,the robot flower sorting system based on deep separation convolutional neural network is deeply studied.The main research work of the thesis is as follows:By reviewing a large number of domestic and foreign literatures,the current research status of convolutional neural networks,the classification of flower images and the application of visual algorithms in automatic robot sorting are summarized and analyzed,and the problems worthy of further study are summarized.Inspired by a large number of optimization algorithms for image classification,it is pointed out that the research on the flower image classification algorithm of micro-embedded devices with limited computing resources from the lightweight convolutional neural network model has important research significance.In view of the software and hardware design of micro embedded devices on the market,the advantages and disadvantages of each scheme are deeply investigated and compared,and the most important part of the development of the device is the choice of the main control CPU chip.It is the best choice for this system to meet both high computing power and low power consumption.It can be used with other peripheral components to meet the company's actual project requirements.In view of the practical application scenarios in the field of industrial flower packaging automation sorting,the image parameters of the flower image classification algorithm are too large,and the sorting precision is not high.It is proposed to use the micro-embedded equipment in this project as the "eye" of the packaging robot.A deep separation convolutional neural network is used to extract the flower features,and the network model structure is analyzed in detail.In order to improve the training speed of the model,a fine-tuned model training method is proposed.The PC-side simulation results show that the flower image classification algorithm is adopted.The accuracy is higher and the stability is better.Finally,the trained model is compressed and optimized,the development environment is set up,the algorithm model is transplanted to the micro embedded device and the classification effect is evaluated.