A Weakly Supervised Object Detection Model for Cyborgs in Smart Cities

With the development of society, the construction of smart cities has become a current research hot spot. The emergence of cyborg has a great impact on the construction of smart cities especially in smart traffic and video surveillance as it can effectively reduce the staffing burden and secure higher reliability. As cyborgs' key technology for maintaining efficient work, object detection can provide localization and classification information, so it is a fundamental task for image understanding and interpretation. Fully supervised object detection has achieved excellent performance, relying mainly on a large number of accurate manual annotated datasets but is time-consuming and costly. Weakly supervised object detection models require only image -level annotation during training, which greatly reduces the annotation cost; thus gradually becoming the mainstream research direction. However, most weakly supervised object detection models tend to fall into the local optimal solution problem when using multi-instance learning, resulting in low detection accuracy. Therefore, we proposed a coordinate attention mechanism to strengthen the feature representation of the relevant regions to alleviate the tendency of models to focus on local objects. A pseudo-label generation branch and a Fast R-CNN branch were designed to optimize our proposed framework jointly. Experimental results on two public datasets showed that the proposed model could effectively alleviate the local optimal solution problem, where the mean accuracy precision and correct localization on the PASCAL VOC 2007 dataset could yield 52.1% and 68.9%, respectively.