Application of an Improved YOLOv8 Model Based on Polarized Images in Trace Detection

With the rapid advancement of deep learning techniques and polarization imaging technology, significant progress has been made in the field of trace detection, particularly in the automatic identification of fingerprints and footprints, is facing increasingly complex environmental challenges. Traditional image processing methods often struggle to accurately identify small trace targets under conditions such as low contrast, complex backgrounds, and varying lighting conditions. To address this issue, this paper presents a trace detection method for polarization images using YOLOv8, combining multi-channel polarization image input, Gabor filters, and the CBAM module to effectively enhance the accuracy and robustness of fingerprint and footprint detection. By utilizing multi-channel polarization image input, the approach capitalizes on the distinctive reflection behavior of polarized light, markedly increasing the contrast between the trace and its surrounding background, while Gabor filters are employed to increase detail features. The introduced CBAM module optimizes the weight distribution of feature maps through an attention mechanism, enabling the model to more accurately capture the features of the target area. The experimental findings clearly indicate that the proposed approach offers a notable enhancement in performance, improving target detection capabilities in complex environments, particularly in fingerprint and footprint detection tasks, while retaining good real-time performance and robustness.