| abstract |
The present invention relates to a kind of night vision image scene recognition method based on deep convolution-deconvolution neural network, it is characterized in that, comprises the following steps: Step 1: build night vision image data set; Step 2: carry out mirror surface to original sample image Symmetrical processing; Step 3: Build a deep convolution-deconvolution neural network; Step 4: Obtain an image of size h×w to be processed in real time, and input the image into the deep convolution-deconvolution neural network to obtain a size of The feature map of h×w; step 5: divide the objects in the night vision image into k different categories, and use the multi-classification algorithm to determine the category of each pixel in the feature map obtained in step 4, and the output size is h×w× Probability plot for k. The invention significantly enhances the scene perception of night vision images, improves the efficiency of target recognition, and reduces the complexity of manual operations. |