【摘要】 冷床区范围较大、结构紧凑、安装金属探测仪困难且成本较高，常用的自动化二级跟踪策略无法有效地完成对该场景下钢板的跟踪。通过利用改进的SOLO实例分割算法构建冷床区钢板检测的视觉模型，识别得到冷床区各个钢板对象的坐标位置和长宽尺寸；再结合逻辑跟踪模型形成多模型匹配跟踪策略，能够识别到行车下料、冷床卡钢等特殊情况，实现自动下料跟踪和异常预警。结果表明，改进的SOLO实例分割方法在6种不同的环境场景下平均mAP(mean average precision)提升到98.14%,平均mIoU(mean intersection over union)提升到98.72%,并且实现了29.1 fps的处理速率，满足了生产条件中对于准确率和实时性的要求。更多还原

【Abstract】 The cooling bed area is large in scope, compact in structure, difficult to install metal detectors and high in cost.The commonly used automatic secondary tracking strategy cannot effectively track the steel plates in this scenario.The visual model of steel plate detection in the cooling bed area is constructed by using the improved SOLO instance segmentation algorithm, and the coordinate position, length and width of each steel plate object in the cooling bed area are identified; Combined with the logic tracking model, a multi model matching tracking strategy is formed, which can identify the special situations such as crane blanking and steel clamping in the cooling bed, and realize automatic blanking tracking and abnormal early warning.The experimental results show that the improved SOLO instance segmentation method can increase the average mAP(mean average precision) to 98.14%,and the average mIoU(mean intersection over union) to 98.72% in six different environmental scenarios, and achieve a processing rate of 29.1 fps, meeting the requirements for accuracy and real-time in production conditions.更多还原