Research on Apple External Quality Grading Method Based on DXNet Model

In the wave of agricultural modernization, intelligent apple selection line technology is a key link in improving fruit quality and production efficiency. There are various methods for grading the external quality of apples, some of which focus on detecting single external features of apples, such as color, size, shape, texture, and visual defects; Other methods are based on deep learning techniques to grade the appearance quality of apples. However, these manual methods for extracting single external features of apples have certain limitations in terms of grading accuracy, and cannot comprehensively and objectively evaluate the external quality of apples. To improve the accuracy and stability of the lossless grading method for apples, this paper proposes a Yan'an apple grading method based on the DXNet model. The convolutional blocks of traditional convolutional neural networks are extracted as feature extractors for apple images, and the learned feature maps are subjected to global max pooling and global average pooling, respectively. The two one-dimensional vectors after pooling are concatenated and input into a classifier mainly composed of fully connected layers for classification.