A Collaborative Framework for Image Enhancement and Feature Extraction Based on the Joint Training Mechanism

In the perception system of autonomous driving, traffic sign detection is a core technology to ensure driving safety, and its environmental adaptability faces significant challenges. In order to tackle the problem of severe performance degradation in existing detection models under extreme weather conditions such as fog, rain, and low light, this study proposes an image enhancement and feature extraction collaborative optimization framework based on a joint training mechanism. By constructing a bidirectional task channel, the denoising process and detection task are deeply coupled. First, we designed an image enhancement module with multi-level feature interactions, which uses clear-blurry image pairs to establish regularization constraints on feature maps, effectively improving the noise suppression capability of the backbone network. Furthermore, we proposed an object-guided image denoising loss function optimization strategy. During the loss calculation process, a region-weighting strategy is introduced to strengthen the restoration accuracy of the target region by adjusting region-sensitive weights. The experimental findings indicate that the proposed approach notably improves the model's resilience and target detection performance in adverse visibility conditions, including fog, rain, and snow.