Improved Accuracy of Detection of Physical Defects in Cocoa Beans with Multi-Scale Wiener Filter on Conveyor Machine

Abstract

Cocoa is an important export commodity that contributes significantly to the global food industry. Quality assessment of cocoa beans, particularly in distinguishing whole beans from broken beans, is generally still done manually, which is subjective, time-consuming, and prone to error. This study proposes a computer vision-based detection system utilizing the YOLOv11 algorithm combined with the Multi-Scale Wiener Filter method to improve detection accuracy in images that are blurred due to conveyor movement. The research dataset comprises 1,338 images of cocoa beans divided into two classes, namely whole beans and broken beans, with three conveyor speed variations (0.60, 0.65, and 0.70 meters per second). The results of this study show that the integration of the Multi-Scale Wiener Filter in YOLOv11 can significantly improve performance compared to the YOLOv8 and YOLOv11 models without filters. At a speed of 0.60 meters per second, the model produced a precision of 0.986, a recall of 0.985, and an mAP at 50 of 0.979. In addition, the model continues to show stable performance at higher conveyor speeds. Overall, this method is effective in reducing the blur effect and improving the accuracy of physical defect detection in cocoa beans, making it potentially applicable in industrial systems.

Access to Document

Other files and links

• Link to publication in Scopus

Fingerprint