Research on skin quality detection system based on facial key point detection

Aiming at the problem that the sampling parts need to be manually marked in the process of facial skin quality detection, a skin quality detection system based on the detection of key points of the face is proposed to automatically locate the sampling parts. In order to realize real-time and fast detection of face key points on portable devices, the system adopts an improved lightweight MobileNetV2 network. Adjusted the model structure to make the model parameters smaller to improve the detection rate. At the same time, in order to improve the detection accuracy of the network, an efficient channel attention (ECA) module is added on the basis of the inverted residual block, which solves the problem of information loss after deep convolution, and can accurately detect face rotation, occlusion, etc. Feature points. Automatic localization is achieved by projecting key points to the face, so as to detect the skin quality of the sampled part. The trained model is tested on the WFLW dataset, and the average error rate of face key points is 1.54%. The experimental results show that the improved network adopted by the system further improves the detection accuracy, which effectively realizes the automatic positioning of the sampling parts in the facial skin quality detection system.