| Title |
Non-destructive Detection of External Fruit Defects using SG Filter·Spectral Differencing of Hyperspectral Data and Recursive Learnging |
| Authors |
김재혁(Jaehyuk Kim) ; 유선호(Seonho Yoo) ; 장숙현(Sookhyeon Chang) ; 김선영(Sunyoung Kim) |
| DOI |
https://doi.org/10.5573/ieie.2025.62.4.84 |
| Keywords |
Fruit quality detection; Hyper spectral imaging; ML; Non-destructive inspection; Differencing |
| Abstract |
A non-destructive inspection method for evaluating fruit quality is essential for the digital transformation of agricultural distribution, enhancing product value, and meeting consumer demands for safe food. Traditional RGB-based inspection methods have limitations in accurately detecting external defects on fruits. In contrast, hyperspectral-based inspection methods can acquire both optical image and spectral characteristics, allowing more precise detection of fruit condition changes. However, existing hyperspectral approaches for detecting external fruit defects?often based on PLSR?average the entire fruit, potentially overlooking localized defects. This paper proposes a method that simultaneously utilizes spectral characteristics and image features to address these limitations. Through a performance analysis combining eight candidate ML methods with SG filtering and spectral difference preprocessing, the XGBoost model demonstrated the highest performance. Additionally, a recursive learning method using pixel-based GT mask labeling for identifying external defects was applied. Compared to the conventional PLSR approach, the proposed method improves accuracy by 27.74% and can detect not only the presence of defects but also the extent of those defects, achieving a high detection accuracy of 98.98%. A hyperspectral fruit-quality testbed was built to collect training data, and 1,000 labeled images were obtained for apples, pears, citrus, and tomatoes, with an aim to develop a low-complexity AI model suitable for real-time operation. The proposed method offers a viable automated solution for commercial fruit grading systems. |