Abstract: The effective surface defects detection of strips is of great importance for ensuring product quality. However, due to low contrast and small target size, existing detection methods often face challenges in achieving sufficient detection accuracy. Therefore, a strip surface defect detection algorithm based on semantic enhancement and local attention mechanism (Scale Aware and Local Attention Detection, SALADet) is proposed. Firstly, a semantic interaction enhancement module is embedded in the backbone network to extract and enhance high-level semantic information in deep feature maps, improving the network's ability to distinguish between background and defects. Secondly, a local attention pyramid is introduced in the neck structure of the network to enhance the feature extraction of small targets, thereby improving the detection accuracy of small-scale objects. To further enhance detection performance, SALADet algorithm employs a decoupled detection head, effectively alleviating the conflict between classification and regression tasks, thus improving overall detection accuracy. Experimental results on the NEU-DET dataset show that the mean average precision of SALADet algorithm reaches 79.4%, representing improvements of 4.7%, 14.1%, 4.5%, 4.6%, and 6.1% over Faster R-CNN, SSD, YOLOX, YOLOv8 and CenterNet algorithms, respectively. Additionally, SALADet algorithm achieves an inference speed of 84.7 frames per second, demonstrating excellent real-time performance and practicality.