X-ray Security Inspection Image Dangerous Goods Detection Algorithm Based on Improved YOLOv4

Aiming at the problems of multi-scale and serious overlap dangerous goods in X-ray security-inspection-image samples, an dangerous-goods-detection algorithm with high detection accuracy is designed based on improvement YOLOv4. Using deformable convolution to redesign YOLOv4’s path-aggregation-network (PANet) module, can flexibly change its receptive field shape detected object. When high-level information low-level are fused PANet used align features, which effectively improve accuracy. Then, Focal-EIOU loss function introduced, solve problem CIOU being prone causing severe loss-value oscillation when dealing low-quality samples. During training, network converge more quickly be slightly improved. Finally, Soft-NMS was non-maximum suppression YOLOv4, solving rate hazardous materials security-inspection dataset improving On SIXRay dataset, this model 95.73%, 83.00%, 82.95%, 85.13%, 80.74% AP for guns, knives, wrenches, pliers, scissors, respectively, mAP reached 85.51%. The proposed reduce false-detection security images ability small targets.