Dynamic Approach for Object Detection using Deep Reinforcement Learning

In this paper, an approach has been proposed for detecting objects using deep reinforcement learning. it involves training an agent to interact with its environment and learn a policy for object detection that maximizes a reward signal. A deep Q-learning algorithm has been used to train the agent to detect objects in images. The agent receives a reward for correctly identifying objects and a penalty for incorrect. For detecting an object, a dynamic method has been tried on the PASCAL VOC 2012 data set. The results demonstrate that this approach reaches a comparable performance level with less labeled data.