欢迎访问西安知识产权运营服务平台
摘要
【 中文摘要 】
本申请属于缺陷检测技术领域，公开了一种基于MSFE‑YOLO的禽类种蛋外壳缺陷检测方法及系统，通过图像采集、光照增强、数据传输以及缺陷检测构成完整的检测系统，实现从图像获取到缺陷识别的一体化检测，在种蛋外壳缺陷检测应用中可有效代替现有的人工检测方法，提高种蛋质量筛选的自动化水平；同时改进MSFE‑YOLO网络模型中的ADEM模块降低了通道维数，加强对高频特征的关注，其次基于PAN‑FPN架构调整特征融合网络，增加对浅层特征的关注，同时引入大尺度卷积核增强对目标上下文信息的关注，最后通过引入可调参数N，实现对IoU的针对性调整，以适应不同检测任务的需求，在检测精度与计算开销之间取得了较好的平衡。
【 英文摘要 】
This application belongs to the field of defect detection technology A method and system for detecting shell defects of poultry hatching eggs based on MSFE ‑ YOLO are disclosed Through image acquisition, illumination enhancement, data transmission and defect detection, a complete detection system is formed to realize integrated detection from image acquisition to defect identification. It can effectively replace the existing manual detection method in the application of egg shell defect detection, and improve the quality of eggs. Automation level of screening; at the same time, improving the ADEM module in the MSFE ‑ YOLO network model reduces the channel dimension Strengthen attention to high-frequency characteristics, secondly, the feature fusion network is adjusted based on the PAN ‑ FPN architecture to increase attention to shallow features. At the same time, a large-scale convolution kernel is introduced to enhance attention to target context information. Finally, the adjustable parameter N is introduced to achieve targeted adjustment of IoU. To adapt to the needs of different detection tasks, a good balance is achieved between detection accuracy and computational overhead.
分类号
其他著录项
| 代理机构 | 陕西瑧萃专利代理事务所(普通合伙) 61314 |
| 代理人 | 刘坤 |
| 申请语言 | 汉语 |
| 审查员 | 鲁思阳 |
其中：
表示CIoU的边界框损失；其中ρ2(b,bgt)表示预测框和真实框中心点之间的欧几里得距离的平方，用于衡量两个框中心点的位置差异；a作为平衡因子用于调整形状相似惩罚项的权重；v表示预测框和真实框的长宽比差异的度量；所述N‑CIoU所提出的交并比损失NIoU表达式为：
其中：