Urban Change Detection based Convolutional Auto Encoder - Content Loss using High-Resolution Images

The Change Detection (CD) in high-resolution images is significant for understanding the surface of land. There have developed numerous Deep Learning (DL) approaches to CD, many of these algorithms failed to predict edges and protect changed area of shape from the multi-spectral images. In this manuscript, the Convolutional Auto encoder - Content Loss (CAE-CL) approach is developed to detect the changes in urban areas. The CL function is used in the CAE, which enhances the performance of CAE to detect the changes of urban areas. Initially, the deep features are extracted by using the initial layers of CAE and then the features are given to final detection layer. This last layer detects the changes in urban areas effectively with high performance. Various evaluation measures considered to evaluate the performance of CAE-CL approach. The CAE-CL technique reached out the accuracy of 99.51%, precision of 99.75%, f1-score of 98.92%, Intersection of Union (IoU) of 98.03% and Kappa Coefficient (KC) of 89.27% while comparing with existing techniques.