Improved CNN non-reference image quality evaluation method

The invention relates to the field of image processing, in particular to an improved CNN non-reference image quality evaluation method, which realizes image quality evaluation based on an improved CNN model, and is characterized in that the improved CNN model is composed of a plurality of convolution, pooling, activation and full connection layers, and the pooling is performed in a maximum, minimum and mean combined mode in the last pooling layer. Therefore, the problem of image feature information loss is solved. According to the method, the thought of transfer learning and fine adjustment is utilized, a single pooling method is replaced in a multi-pooling combined mode, the performance of the network is improved, the accuracy is high, the actual requirement of current no-reference image quality evaluation can be met, and the experimental result shows that the method is suitable for large-scale popularization and application. The evaluation accuracy of the method on a standard image quality evaluation librar