Endoscopic Lesion Detection Based on Improved SSD

Aiming at the problem of low accuracy and speed of lesion detection by lightweight model, an endoscopic lesion detection algorithm based on improved SSD is proposed. Firstly, the VGG backbone network is replaced with a lighter MobileNetV2 network to reduce the consumption of computing resources. Secondly, the CBAM attention mechanism is proposed to refine the shallow input features, so as to comprehensively capture the key information of the features. Finally, the improved SmoothL1 loss function is used to balance the accuracy and robustness of the model. Experiments show that the average accuracy of the improved model is improved, the parameter amount is 1/6 of the original model, the calculation amount GFLOPS is reduced by 59 G, and the frame rate per second FPS is increased by 96 frames/s. Compared to other algorithms, there is an improvement in average accuracy, the model is more lightweight, and it consumes fewer computing resources, indicating the effectiveness of the improved algorithm in the detection of endoscopic lesions.