In the infrared target detection methods based on temperature difference, the final target detection result is greatly influenced by the segmentation effect of the infrared image.As to infrared images with certain detection distance, the traditional segmentation method is not fit for the segmentation of the target edge.To solve the problem, Convolutional Neural Network (CNN) is used to extract image features.A spectral-clustering infrared image segmentation method based on CNN is proposed.Experimental results show that this method can improve the segmentation accuracy of infrared images.Compared with that of the traditional image segmentation method based on sparse matrix and spectral clustering, the clustering speed of the CNN-based spectral-clustering infrared image segmentation method is increased by about 16 times.Under certain detection distance, the improved spectral-clustering infrared image segmentation method can acquire more accurate results of the temperature difference between the target and the background.