SRM-FPN: A Small Target Detection Method Based on FPN Optimized Feature

With the development of network detection models, researchers have achieved good results in general target detection, but there is still no good solution for small target detection in images, especially the feature processing of small targets. At present, the most suitable feature processing method is FPN, but this fusion method will cause the feature redundancy, ambiguity and inaccuracy of small targets, and has little effect on the general large targets, but it will cause great interference and errors in the detection of small targets. For the above problems, this paper improves FPN and proposes a new SRM-FPN feature fusion method. Specifically, SRM is a spatial refinement model that learns the location of future feature points according to the context features between adjacent layers and content, while borrowing the adaptive channel merging method of the attention mechanism to optimize feature fusion. Compared with other methods, the optimized scheme combined with the existing model can effectively improve the detection effect of small targets in the image.