Adaptive Kuan Regressive Gene Optimized Feature Selection based Tucker’s Congruence Deep Convolutional Learning for Change Detection using Satellite Images

Change detection in multi-temporal images is a remote sensing application detects land cover changes that occurred between two satellite images acquired at different times in same geographical region but different obtained from different types of sensors. Several research works have been conducted in change detection but accurate detection with minimum time still remains a challenging issue. A novel technique called Adaptive Kuan Regressive gene optimized feature selection-based Tucker’s Congruence Deep Convolutional learning (AKRGOFS-TCDCL) is proposed for accurate change detection with minimum time. The proposed AKRGOFS-TCDCL technique involves three processes namely preprocessing, feature selection, and classification. Preprocessing of atmospheric corrections, radiometric correction, topographic correction, and contrast enhancement are performed using Adaptive Kuan filtering. With the preprocessed image, optimal features are selected by means of machine learning-based GA called Dichtomous probit Regression, for minimizing time consumption. Finally, classification is performed using Tucker’s congruence coefficient deep convolutional neural learning for detecting changes in given satellite images via feature matching. In this way, accurate change detection is performed with minimum error. An experimental evaluation of the proposed AKRGOFS-TCDCL technique and existing methods are performed using satellite image dataset. The results are discussed with different performance metrics such as detection rate, false-positive rate, and detection time with respect to different satellite images.