Food Safety Inspection Based on Deep Learning

This study focuses on the intelligent transformation in the field of food safety testing, aiming to address the challenges in meat product quality assessment by integrating deep learning technology and image processing methods. This study aims to improve the accuracy and efficiency of detection by using Convolutional Neural Networks (CNNs) and Dense Connected Networks (DenseNets) to classify pork images into safe and unsafe categories. Through rigorous image preprocessing, including bilateral filtering and data augmentation techniques, the quality of the training data is ensured. The model training and validation adopted an 8:2 training test dataset partitioning ratio to enhance the model's generalization ability and prediction accuracy. For DenseNet, we fully utilize its feature reuse and propagation capabilities, optimize the network structure and connection methods to ensure that the model can efficiently extract and utilize image features. The experimental results show that the DenseNet model achieved an accuracy of 73.8% in meat image classification tasks, verifying its effectiveness and reliability in the field of food safety monitoring.