Research of Image Recognition and Classification Based on NIN Model
Mengxi Liu, Yongfeng Ju, Jiuxu Song, Zheng Wang · Journal of Physics Conference Series · 2018
Focus on the weak feature expression ability of traditional Convolution Neural Network (CNN) in image recognition and classification, the CNN has been improved and optimized by introduction of Mlpconv layer to form a Network in Network (NIN). The NIN enhances feature extraction and abstract expression of the local patch; get better performance of the image recognition. The experiments of weld image recognition and classification show that the optimized CNN can improve the feature expression ability of the whole network by reducing the number of parameters, obtain higher recognition and classification precision, and avoid the fitting of the network model effectively. The CNN achieve improvement in both performance and efficiency.