Apple Defects Detection Using Principal Component Features of Multispectral Reflectance Imaging

Apple defects detection using hyperspectral imaging has become an active research topic during the last decade. The main merit of hyperspectral imaging is that it has a lot of information, but its data size is big. In the hyperspectral imaging, the most challenging aspect is to reduce the data size while keeping the vital information. Small data size is an essential component for real-time processing that the industries need. The methods to reduce the data size for hyperspectral imaging are generally statistical methods. In this paper, the statistical data reducing method is specialized for apple hyperspectral image data. This paper proposes an apple defects detection using multispectral imaging and principal component analysis. In the preprocessing, we examine the image quality for all the hyperspectral apple image in the spectral range from 403 to 988 nm and select three wavelengths. In the main processing, we perform principal component analysis for the three wavelengths and choose the best principal components for apple defects detection. And the defected apples are detected sequentially using the principal components and global thresholds. We show the algorithm for the above processing and an experiment with hyperspectral apple images. Preliminary examination shows that the general detection rate is 97%.