Synchronous detection of internal and external defects of citrus by structured-illumination reflectance imaging coupling with improved YOLO v7

In real growth environments, citrus fruit defects exhibit complex and diverse, encompassing both external defects (e.g., flavedo disorder, cracking, canker, wind scarring) and internal defect (e.g., early decay). In addition, the stem and navel of the fruit can also be misjudged as defects. To address the limitations of traditional methods in simultaneously identifying external and internal defects in citrus, this study proposes a citrus defect detection method that combines structured-illumination reflectance imaging (SIRI) technology with an improved YOLO v7. Original stripe images with a spatial frequency of 0.25 cycles mm⁻¹ were collected under a light-emitting diode (LED) SIRI system. The three-phase-shifting approach was employed to recover the direct component (DC) and alternating component (AC) images. The AC image contains tissue information at a specific depth, enabling the simultaneous visualization of early decayed areas that are imperceptible to the naked eye and surface defects on the sample. This provides a foundation for the synchronous detection of both external and internal defects in citrus. Using AC images as input and YOLO v7 as the recognition model, the external and internal defects in citrus were identified synchronously. The mAP_0.5 and F1 of YOLO v7 were 93.5 % and 92.4 %, respectively. Compared to the original YOLO v7, gradually replacing the CA module, Mish module, and SPPFS module and conducting ablation experiments, resulting in an increase of 1.4 %, 0.9 %, 0.7 %, and 2.4 % in mAP_0.5, as well as an increase of 1.9, 1.2, 0.6, and 2.7 in F1, respectively. The improved YOLO v7 increased the accuracy of identifying both external and internal defects. Using an independent batch of samples and the trained YOLO v7-CA-Mish-SPPFS detection model, the overall classification accuracy achieved 94.9 %. The results demonstrate that the combination of SIRI technology with the improved YOLO v7 model can effectively identify external and internal defects in citrus fruit, thereby reducing duplicate detection of samples. This study provides a new approach for the first synchronous detection of external and internal defects in the citrus industry.