A novel image detection method for internal cracks in corn seeds in an industrial inspection line

文献类型： 外文期刊

第一作者： Zhang, Yuzhuo

作者： Zhang, Yuzhuo;Wang, Decheng;Lv, Chengxu;Mao, Wenhua;Li, Jia

作者机构：

关键词： Random falling; Uniform direction; S2ANet algorithm; Rotating target detection; Edge detection operator

期刊名称：COMPUTERS AND ELECTRONICS IN AGRICULTURE （ 影响因子：6.757； 五年影响因子：6.817 ）

ISSN： 0168-1699

年卷期： 2022 年 197 卷

页码：

收录情况： SCI

摘要： Internal cracks produced on the inner endosperm side of corn seeds are difficult to identify. During actual production, the random falling of numerous corn seeds increases the difficulty of crack detection. To date, there has been little research on internal crack detection in corn seeds. This paper reports a new detection method that combines a deep learning algorithm with edge detection threshold processing. This method is based on the optimized S2ANet model. The Average Precision (AP) value of the optimized S2ANet model reached 95.6%, >9% higher than that of the original model. This method also involves novel methodological approaches for obtaining uniform seed direction. Experiments showed that it effectively solves the problems of random seed numbers, random seed orientations, different types of light source transmission, and difficulties in dataset production. In an experiment with individual seed types (cracked and non-cracked), the identification accuracy was 95.08% for cracked seeds and 95.75% for non-cracked seeds. In an experiment with a mixture of seed types, the average precision of 10 trials was 95.91%, the average recall was 94.8%, and the average F-score value (F1) was 95.34%. This level of accuracy is higher than that achieved through the direct use of deep learning detection algorithms. This study makes a valuable contribution to the detection of internal cracks in corn seeds on industrial inspection lines. At the same time, it provides a new approach for non-destructive testing of plant seeds.

分类号：

  • 相关文献
作者其他论文 更多>>

微信小程序