HSV 색공간을 활용한 실시간 커피생두 불량 검출 시스템

초록

The purpose of this paper is to present a defect detection system utilizing the HSV color space for the visual quality inspection of green coffee beans. Two algorithms were applied in this experiment. The first involves imaging objects received from a line scan camera and separating each object without duplication. The second entails obtaining the HSV distribution of the objects in the form of graphs and numerical data. The comparative subjects are normal beans and defective beans. Coffee can vary depending on cultivation and storage conditions, with intact beans referred to as normal beans, while defective beans refer to those that have deteriorated in quality due to poor cultivation conditions, deformities due to improper sprouting, or contamination by mold, pests, or foreign substances. The H, S, and V values are established as reference values based on 40-50% of the green coffee beans. When processing the video frames from the line scan camera into a single image, issues can arise, such as object truncation at the image boundaries or the same object being detected again in subsequent images. To address these issues, a feature called 'HB (Historic Buffer)' was added to the input section to recognize and process object truncation. The beans were divided into three groups: normal beans, defective beans A (with 75-100% defects), and defective beans B (with 50-75% defects). The average H, S, and V values of each group were compared, confirming that it is possible to distinguish between normal and defective beans based on value differences.

키워드