Task-aligned Object Detection Algorithm Based on Random Scaling Mixture and Cross-Scale Feature Enhancement
Aimed at the disadvantages of poor robustness,some semantic information loss in the top-level feature layer of feature pyramid network in the TOOD algorithm,and existing a semantic gap for feature layers with different scales,a task-aligned object de-tection algorithm based on random scaling mixture and cross-scale feature enhancement is proposed.The 2×4 hybrid augmentation method is used to enrich the training samples,and improve the generalization and robustness of the model.The multiple residual fea-ture enhancement module is constructed to adaptively fuse the context information with different scales at the top level,and reduce the semantic information loss at the highest level.Moreover,the proposed method constructs the stacked pyramid convolution module,reduces the semantic gap among the different scale features,and improves the effect of multi-scale feature fusion.Experimental re-sults on the Pascal VOC data set show that the mean average precision,precision,and recall of the proposed algorithm are 3.76％,15.71％and 6.28％respectively higher than that of the TOOD algorithm.The presented algorithm is superior to six mainstream comparison algorithms in the F1 value and mean average precision.
万方数据
维普
