An Aircraft Detection Framework Based on Reinforcement Learning and Convolutional Neural Networks in Remote Sensing Images

Abstract

Aircraft detection has attracted increasing attention in the field of remote sensing image analysis. Complex background, illumination change and variations of aircraft kind and size in remote sensing images make the task challenging. In our work, we propose an effective aircraft detection framework based on reinforcement learning and a convolutional neural network (CNN) model. Aircraft in remote sensing images can be accurately and robustly located with the help of the searching mechanism that the candidate region is dynamically reduced to the correct location of aircraft, which is implemented through reinforcement learning. The detection framework overcomes the difficulties that the current detection methods based on reinforcement learning are only able to detect a fixed number of objects. Specifically, we adopt the restricted EdgeBoxes that generate the high-quality candidate boxes through the prior aircraft knowledge at first. Then, we train an intelligent detection agent through reinforcement learning and apprenticeship learning. The detection agent accurately locates the aircraft in the candidate boxes within several actions, and it even performs better than the greed strategy in apprenticeship learning. During the final detection step, we carefully design the CNN model that predicts the probability that the localization result generated by the detection agent is an aircraft. Comparative experiments demonstrate the accuracy and efficiency of our aircraft detection framework.

Deep Learning for Change Detection in Remote Sensing Images: Comprehensive Review and Meta-Analysis

“…Fu et al [122] have shown the feasibility of using deep reinforcement learning for remote sensing ship detection task. Recently, Li et al [123] have proposed an interesting aircraft detection framework based on the combination of a CNN model with reinforcement learning. Similarly, the change detection process can be solved as an action-decision problem based on a sequence of actions refining the size of the changed regions between two input images.…”

Adaptive Human–Machine Evaluation Framework Using Stochastic Gradient Descent-Based Reinforcement Learning for Dynamic Competing Network

“…The route planning problem is often addressed in RL, utilizing an RL algorithm that yields the maximum reward value for an unmanned ship. In aircraft detection systems and radar designs, RL is applied to optimal radar system design and aircraft image analysis to detect radio waves and minimize unnecessary interference [25][26][27].…”

Remote sensing image description based on word embedding and end-to-end deep learning

“…The approach not only considered the complex spatial dependence of labels but also effectively learned the textural and spatial characteristics of images. Li 13 proposed a deep image detection and description framework by combining enhanced learning with a deep CNN to address the increased difficulty of aircraft identification in remote sensing images owing to a series of problems such as illumination and changes in aircraft type and size. The approach was demonstrated to accurately identify the specific positions of aircraft in remote sensing images.…”