Object recognition system of sports equipment based on convolutional neural network
Ying-Chen · 2023
Based on the performance characteristics of DSP and convolutional neural network technology, this paper tries to solve the shortcomings of traditional image recognition technology such as slow speed, low precision and high equipment requirements, and realizes the operation of convolutional neural network framework on embedded devices. According to the principle of similar scale, these public image data are screened and classified, and individual images that affect the classification are eliminated. The data set is expanded by reflection and noise, etc., and the effect is tested by experimental network training. The final data set is formed (the training set contains 53,000 pictures, and the test set contains 21000 pictures). Understand the performance of different network frameworks and network models in the field of image recognition, combined with the characteristics of embedded devices to be used, determine the appropriate network for model training. The transfer learning strategy is adopted to improve the training speed and efficiency, and finally a more effective recognition model is obtained.