Multi Feature Apple Grading Based on Entropy Method and MLP

A multi feature apple grading method based on entropy method and MLP is proposed to address the inaccurate calculation of grading indicators and the problem of multi feature fusion caused by surface reflection in Aksu Red Fuji apple grading. In preprocessing, grayscale is applied to the collected image, threshold segmentation is performed to obtain the reflective area, morphological operations are performed to remove noise, and Canny detection and non-maximum suppression are used to extract the edge contour of the reflective area. The reflective area is filled with white, and binarization is used to fuse with the original image, adjusting the brightness channel image value. By using the entropy method and manual selection preferences to simulate real grading scenarios, the weights of five features of apples, including size, shape, color, defects, and texture, are extracted. Combined with manual selection preferences, the proportion of each feature weight in apple grading is comprehensively obtained. A MLP based apple grading method is proposed based on weighted features, with a grading accuracy of 98%, verifying the feasibility of the proposed method and providing theoretical reference for the grading of Red Fuji apples.