Night vision image scene identification method based on deep convolution-deconvolution neural network

Abstract

The invention relates to a night vision image scene identification method based on a deep convolution-deconvolution neural network. The method is characterized in comprising the steps of S1, establishing a night vision image data set; S2, carrying out mirror symmetry processing on original sample images; S3, establishing the deep convolution-deconvolution neural network; S4, obtaining to-be-processed images with sizes of h*w in real time and inputting the images into the deep convolution-deconvolution neural network, thereby obtaining feature graphs with the sizes of h*w; and S5, dividing the objects in the night vision images into k different classes, determining the class to which each pixel in the feature graphs obtained in the S4 belongs through adoption of a multi-classification algorithm and outputting probability graphs with the sizes of h*w*k. According to the method, the scene sensation of the night vision images is clearly increased, the target identification efficiency is improved, and the manual operation complexity is reduced.

Description

technical field [0001] The invention relates to a night vision image scene recognition method based on a deep convolution-deconvolution neural network, belonging to the field of night vision image processing. Background technique [0002] Scene recognition refers to identifying the scene in the picture according to the similar content of the scene image. Scene recognition is a basic preprocessing process in the fields of computer vision and robotics, and it plays an important role in computer intelligence fields such as image content retrieval, pattern recognition, and machine learning. [0003] Scene recognition technologies mainly include scene recognition methods based on object recognition, scene recognition methods based on image area recognition, scene recognition methods based on context analysis, and scene recognition methods that imitate biological vision mechanisms. [0004] In the field of visible light color images, the research on scene recognition has made gre...

AI Technical Summary

Problems solved by technology

However, the disadvantage of this algorithm is that it has very high requirements for the establishment of the sample library in the early stage, and the selected GIST features are not sensitive to the shape, category or specific position of the object in the image.

Images