In order to obtain useful information from infrared spectral images,an adaptive segmentation method based on improved transfer learning is designed.The acquisition system obtains the infrared spectral image,uses histogram equalization technology to en-hance the infrared spectral image,reconstructs the infrared spectral image according to the Gaussian normal distribution of each frame pixel of the enhanced image,inputs the infrared spectral image reconstruction results into the two branch convolution neu-ral network model,uses the migration learning subnet to extract the global characteristics of the infrared spectral image,and com-bines the small features of the image extracted by the residual attention subnet.Through integrated learning and fusion of differ-ent infrared spectral image features,the segmented infrared spectral image is output.The test results show that this method can achieve infrared spectral image segmentation,and the segmentation results are clear,with significant details.
improved transfer learninginfrared spectral imageadaptive segmentationdouble branched convolutionresidual attentionthe spread of smooth
万方数据
