Visual quality assessment of plant-based meat products using deep learning

Abstract

The growing demand for plant-based meat alternatives has heightened the need for robust quality control systems to ensure visual and structural consistency. Traditional quality assessment methods rely heavily on human inspection, which can be subjective, time-consuming, and prone to variability. This thesis investigates the use of deep learning techniques, particularly convolutional neural networks (CNNs), to automate the visual quality assessment of plant-based meat products. A modified ResNet-18 model, pre-trained on ImageNet, was fine-tuned to perform regression on expert-provided quality scores. The dataset consists of high-resolution images labeled by human evaluators, covering a range of visual characteristics such as color, surface texture, and structural uniformity. Data preprocessing included resizing, normalization, and augmentation to improve model robustness. The model was trained using a combination of optimization techniques, including the Adam optimizer, learning rate scheduling, and regularization strategies like dropout and weight decay. Performance was evaluated using standard regression metrics, including Mean Squared Error (MSE), Mean Absolute Error (MAE), and R². The overall effectiveness of the proposed convolutional neural network (CNN) model was demonstrated in terms of all these measures, suggesting that deep learning can be effectively applied to quality assessment tasks in plant-based meat production. This work demonstrates the feasibility of using deep learning for visual quality assessment in food production. By automating the evaluation process, the proposed approach has the potential to improve consistency, reduce labor costs, and support scalable quality control in the plant-based food industry.