Lightweight research based on FCOS steel defect detection

At present, the state has put forward a high demand for quality control of the steel produced. Steel defect detection plays a crucial role, and most of the research now not only needs to set the prior frame in advance, but also needs to manually adjust the parameters of the anchor frame, so there are many redundant boxes, resulting in a waste of resources. Therefore, based on the Full Convolution One-Stage Object Detection (FCOS) algorithm without anchor frames, this study proposes a lightweight object detection model MmS-FCOS. For FCOS, the original backbone network ResNet50 is replaced with MobileNetV2, the depth separable convolution is used to replace the ordinary convolution, and the inverted residual structure is added, which not only greatly reduces the number of model parameters and calculations, but also improves the accuracy of algorithm detection AP by $5.6 \%$, and improves the detection accuracy of small, medium and large targets by $3.3 \%, 3.6 \%$ and $7.1 \%$ respectivelyAnd experiments verify the improvement of mSE on the accuracy of the whole network.