Surface Defects Detection of Transparent Plastic Bottles Based on Improved Yolov5

Given the high demand for transparent plastic bottles in our daily lives, coupled with their appearance being a crucial quality indicator, it is imperative to pay close attention to detecting surface defects on these bottles.Traditional manual detection methods are inefficient, time-consuming, labor-intensive, and costly. A deep learning technique is proposed in this work to perform automatic transparent plastic bottles surface defects detection by improving a YOLOv5 object detection algorithm. To enhance the precision of small target detection in YOLOv5, we propose a more friendly pyramid pooling model for small target(SOSPP). This model replaces the pooling operation with dilated convolutions featuring different expansion rates and adds a small target branch.At the same time,it incorporates an ECA and CBMA attention mechanism to improve object recognition for small objects. Moreover, the multi-scale pyramid channel spatial fusion attention mechanism(PCSA) has been incorporated into the feature extraction stage to enhance the detection of small targets.The effectiveness is evaluated based on our exclusive dataset, and the outcomes indicate that the enhanced YOLOv5 significantly improves its ability to detect small target defects.