基于ZC-YOLO的棉花杂质检测.

王中璞; 吴正香; 张立杰; 阿不都热西提•买买提; 张 倩

Aiming at the problems of low detection accuracy caused, by complex shape and large scale change of cotton impurities, an intelligent classification detection method based On improved YOLOvS was proposed. The adaptive anchor frame algorithm adapted, to the data set was used, to recluster the anchor frame to improve the detection effect of small target impurities* In the feature fusion part, MCA attention mechanism module was introduced to focus the impurity target information of the effective feature layer reduce the interference of irrelevant areas and locate the cotton impurity target more accurately. The GIoU loss function was used to calculate the loss of cotton impurity prediction box and real box, and the best cotton impurity detection box was filtered. Out, which makes the algorithm more suitable for the current detection task* Experimental results show that the average accuracy of the proposed algorithm model (mAP@ 0. 5) reaches 92. 5%. Compared with YOLOv3, YOLOv5, YOLOvS and YOLOv6, the mAP index of the proposed algorithm was improved by 15. 4%, 2. 2%, 13. 5% and 26. 4% respectively* It provides reference for intelligent classification and detection of cotton impurities, and the accuracy of model detection is improved.

COTTON; CLASSIFICATION; ALGORITHMS; FORECASTING

Wool Textile Journal, 2024, Vol 52, Issue 12, p95

1003-1456

Academic Journal

10.19333/j.mfkj.20240505707