Detection method for infrared thermal image damage area of coal rock

A detection method for an infrared thermal image damage area of a coal rock, comprising the following steps:

• step 1:
• collecting and recording an infrared thermal image of the coal rock in a process of loading destruction;
• step 2:
• processing a gray-scale transformation on the infrared thermal image of the coal rock;
• step 3:
• denoising the infrared thermal image of the coal rock after the gray-scale transformation by a dense residual image denoising algorithm of an autocorrelation network;
• step 4:
• conducting an area segmentation on the infrared thermal image of the coal rock after denoising to extract eigenvalues by using a coal-rock infrared thermal image damage area segmentation algorithm of an improved encoder-decoder network;
• and
• step 5:
• observing a damage area of the coal rock;
• wherein the dense residual image denoising algorithm of the autocorrelation network in step 3 comprises an asymmetric multi-scale convolution module, a dense residual cascaded autocorrelation block and a reconstruction module, and a global residual connection and a skip connection are also introduced;
• wherein a specific process of the dense residual image denoising algorithm of the autocorrelation network in step 3 comprises:
• S3-1, extracting shallow features from a coal-rock infrared thermal image with noise through the asymmetric multi-scale convolution module, and extracting accurate detailed features in the infrared thermal image of the coal rock;
• S3-2, using the dense residual cascaded autocorrelation block to further extract deep features;
• S3-3, splicing the shallow features extracted by the asymmetric multi-scale convolution module with the deep features extracted by the dense residual cascaded autocorrelation block by using the skip connection;
• and
• S3-4, using the reconstruction module to reconstruct a network to obtain a denoised image, during reconstruction, using a 3×3 convolutional layer to adaptively adjust fusion features, and introducing the global residual connection to learn residual information between images;
• wherein a specific process of the coal-rock infrared thermal image damage area segmentation algorithm of the improved encoder-decoder network in step 4 comprises:
• S4-1, in an encoder, inputting an image with a size of 256×256 and a number of channels of 1, after feature extraction by two 3×3 convolutions, using a 2×2 maximum pooling to downsample and compress the features, after four repetitions, inputting a feature map with rich semantic information at this time into a series empty space pyramid pooling module to further extract multi-scale semantic information of the feature map;
• S4-2, in a decoder, inputting the feature map output by the series empty space pyramid pooling module after sampling and a feature map with a corresponding scale size of the decoder into an attention feature fusion module to replace an original skip connection;
• and
• S4-3, obtaining a feature map with an original size, and transforming a predicted structure obtained by a feature vector mapping into a probability distribution map through 1×1 convolution, wherein the probability distribution map is input into a dense conditional random field to further optimize segmentation results, and the results are output after repeated reasoning.