A Convolutional Neural Network Fuzzy Image Classification Method Based on Image Enhancement

Abstract

The present invention relates to the technical field of image classification, and provides a fuzzy image classification method based on convolutional neural network based on image enhancement. Firstly, the cumulative function of each gray level of the fuzzy image to be classified is calculated, and the mapping relationship of the global histogram is established to obtain the global histogram. Figure the equalized image and calculate its dark channel, atmospheric light components, and transmittance to perform dark channel dehazing processing; then build a cascade network, input the image-enhanced blurred image into the positioning network, and extract it in the feature extraction network Feature map, the candidate frame proposal network generates candidate frames, features are extracted in the ROI pooling layer, and the target and its coordinate information are obtained through the classification and regression of the fully connected layer, and the target is cut accordingly; finally, the target is input into the classification network for feature Extract and classify in the softmax classifier to obtain the category of the target in the fuzzy image to be classified. The invention can improve the accuracy of fuzzy image classification, and reduce the false recognition rate and missed recognition rate.

Description

technical field [0001] The invention relates to the technical field of image classification and recognition, in particular to an image enhancement-based convolutional neural network fuzzy image classification method. Background technique [0002] The main function of image classification and recognition is to locate the target of interest to the input image information, and to classify the target. At present, image classification and recognition technology has been widely used in the field of intelligent traffic management, which can significantly improve the performance of traffic supervision and vehicle control in intelligent traffic management systems. [0003] At present, traditional image classification methods mainly include image recognition based on statistical patterns, image recognition based on pixel features, image recognition based on feature description operators, image recognition based on BP neural network, and image recognition based on fuzzy comprehensive e...

AI Technical Summary

Problems solved by technology

However, none of these image classification methods combine the image enhancement method with the convolutional neural network-based cascaded network and apply it to image classification. When extracting features, the complex background has a relatively large negative impact on the target classification results. When the target is imaged, the image blur caused by the special imaging environment has a relatively large impact on the extracted features. The cascade network target classification method is not clear enough for unclear pictures and features. The target will produce misidentification and missed identification

Images