Low-Light Image Object Detection Algorithm Based on YOLOX

In response to the insufficient image feature extraction and lower detection accuracy of existing object recognition algorithms in low-light environments, this paper proposes a low-light image object detection algorithm, DCEYOLOX, based on an improved YOLOX framework. Firstly, Zero-Reference Deep Curve Estimation is employed for image preprocessing to obtain enhanced images. Subsequently, a dual-channel feature fusion module based on channel attention mechanisms is designed to fuse the enhanced image with the original image features, thereby reducing the impact of amplified noise in the enhanced image and enhancing the recognition capability of blurry objects. Finally, a lightweight MobileNetv3 network replaces the YOLOX-s backbone network, introducing Coordinate Attention to optimize spatial information, achieving a balance between algorithm accuracy and speed.