Face Recognition Model Optimization Research Based on Embedded Platform

The development of information technology has promoted the expansion application field facial recognition technology. Its mainstream methods rely on deep learning algorithms for calculation, but problem large data computation brought by its system makes it difficult to apply in embedded platform devices. As a result, this study focuses improving systems built lightweight backend networks and builds an environment using multi-scale feature fusion, anchor box size optimization, addition channel attention mechanism weighted features, affine face alignment, file compilation. experimental results showed that when number iterations was 300, loss value (0.46) improved algorithm much smaller than other comparison (1.42, 1.73, 2.01), ACC (0.924) significantly better (0.915, 0.909, 0.894). minimum testing time consumption 7 ms. This high accuracy fast running speed, is less limited environmental conditions types.It ideally suited use hardware devices, broadening scope equipment matching algorithms’ applications. satisfy demands devices massive processing jobs.