基于TD-YOLO的靶标与水柱检测

Target and water column detection based on TD-YOLO

海上靶标和水柱位置是用来判定打靶效果的重要信息,实现靶标和水柱自动识别对于实现射击效果快速评估有重要意义。为此,针对靶球目标尺寸小、海上水柱目标与背景高度相似等特点,提出由Backbone、Neck和TD-Head组成的TD-YOLO目标检测模型。为进一步加速模型收敛,模型选用SIOU作为回归框损失函数,并引入剪枝模块,解决了检测过程中出现的水柱检测过拟合现象。对比实验结果表明:TD-YOLO靶标和水柱目标检测模型在靶标和水柱检测数据集上过拟合情况明显减少,检测效果良好。

The position of the target at sea and the water column are important information to determine the shooting effect.It is of great significance to realize the automatic identification of the target and the water column for the rapid evaluation of the shooting effect.In view of the small size of the target ball and the high similarity between the water column target and the background,a TD-YOLO target detection model was proposed,which was consisted of Backbone,Neck,and TD-Head.Meanwhile,in the target and water column detection task,SIOU was selected as the loss function of the regression box,and a pruning module was introduced to solve the over-fitting phenomenon of water column detection in the detection process in order to further accelerate the convergence of the model.Comparative experiments show that the TD-YOLO target and water column target detection models have significantly reduced overfitting on the target and water column detection datasets,confirming the good detection effect.