Three-dimensional visualization and detection of early bruise in apple based on near-infrared hyperspectral imaging coupled with geometrical influence correction

Visible-near-infrared (Vis-NIR) spectral imaging holds great promise for the automatic detection of fruit defects. However, uneven brightness resulting from fruit geometry and the limitations of one-directional imaging significantly restrict the current detection to a limited area. This study presents a rotation measurement system that combines a line-scan NIR-hyperspectral imaging (HSI) camera with a laser profile. A total of 72 apple samples with bruises in the central and edge regions were prepared. A 360° scanning approach was employed to collect HSI and shape data from the entire surface of the samples over a 6-h post-bruising period. Height and angle corrections were applied to eliminate the surface geometric influences on the HSI data, resulting in improved reflectance spectrum uniformity. A two-step principal component analysis method was employed for image enhancement, followed by a straightforward bruise detection technique using global segmentation and connected-domain selection. The results demonstrated an overall improvement in bruise detection over time. Moreover, the correction significantly enhanced the detection accuracy. After 6 h of bruising, the corrected data achieved a classification accuracy of 90.3% and an identification rate of 83.3% for central bruises and 61.1% for edge bruises, whereas the uncorrected data yielded 70.8%, 58.3%, and 31.9%, respectively. Thus, this study successfully detected early bruising across the entire surface of apples and improved the detection in low-intensity edge areas. The proposed method has the potential to contribute to the comprehensive evaluation of agricultural products with irregular geometries.