Remote sensing image detection method in complex scenes based on SPAN and improved BIT

With the acceleration of urbanization, the impact of human activities on the natural environment is becoming increasingly significant, which requires the detection of changes in complex scenes in a timely and scientific manner. This study proposes a remote sensing image change detection model based on spatial pyramid attention network (SPAN) and an improved bit-time image transformer (BIT). The experimental results on the LEVIR-CD dataset revealed that the accuracy, recall, F1 score and overall accuracy of the proposed model reached 0.984, 0.863, 0.795 and 0.850, respectively. Compared with the early fusion method of full convolution, these four evaluation indicators are improved by 8%, 5.2%, 21.6% and 10.4%, respectively. The above results indicate that the research method can effectively solve the problem of information loss in multiscale feature fusion and reduce memory usage to 40.52% of the benchmark model while maintaining accuracy.