Structured Object-Level Relational Reasoning CNN-Based Target Detection Algorithm in a Remote Sensing Image

Deep learning technology has been extensively explored by existing methods to improve the performance of target detection in remote sensing images, due its powerful feature extraction and representation abilities. However, these usually focus on interior features target, but ignore exterior semantic information around especially object-level relationship. Consequently, fail detect recognize targets complex background where multiple objects crowd together. To handle this problem, a diversified context fusion framework based convolutional neural network (DCIFF-CNN) is proposed paper, which employs structured relationship recognition backgrounds. The DCIFF-CNN composed two successive sub-networks, i.e., multi-scale local region proposal (MLC-RPN) an (ORC-TDN). MLC-RPN relies fine-grained details generate candidate regions image. Then, ORC-TDN utilizes spatial integrating attentional message integrated module (AMIM) object relational graph (ORSG). AMIM into feed-forward CNN highlight useful information, while ORSG builds relations between set processing their appearance geometric features. Finally, method effectively represents exploiting both multiscale relationships. Extensive experiments are conducted, experimental results demonstrate that improves accuracy backgrounds, showing superiority other state-of-the-art methods.