苹果内在品质在线检测方法及应用OA北大核心CSTPCD

To address the limited application of post-harvest apple quality detection devices,singular evaluation parameters,and low detection accuracy leading to inconsistent apple quality,the hardware components of a robotic system were developed and tested,and a computer vision system for apple target recognition was established based on the YOLOv9 object detection algorithm.In order to facilitate apple grasping and internal quality inspection,an end-effector was designed and manufactured to perform apple grasping and near-infrared spectral acquisition.Various preprocessing methods were applied to build Partial Least Squares Regression(PLS) models for each indicator after preprocessing.The results indicate that the target recognition accuracy for apples reaches 0.9908.Among different spectral preprocessing methods,the combination of Normalized Spectral Ratio(NSR) and Competitive Adaptive Reweighted Sampling(CARS) achieves the best modeling effect for sugar content and hardness,whereas the MSC+CARS combined preprocessing method yields the optimal PLS model for acidity.The correlation coefficient of the calibration set (Rc) and the correlation coefficient of the prediction set (Rp) for the sugar content model are 0.9789 and 0.9769,respectively,with the root mean square error of the calibration set (RMSEC) and the root mean square error of the prediction (RMSEP) of 0.3006％ and 0.3382％,respectively.An independent verification with 40 apples shows that the correlation coefficient of the verification set (Rv) for sugar content is 0.9683,and the root mean square error of the verification set(RMSEV) for sugar content reaches 0.430％.The grading function of the robotic system was validated,and an overall grading accuracy of 95％ was achieved.This study is of important significance for post-harvest sorting and non-destructive testing in related fields.