Deep learning-enhanced multi-modal data fusion for predicting gel strength and water holding capacity in ultrasonically processed pork

Abstract

Improving gel quality in minced pork remains a key challenge in the meat processing industry, affecting both product integrity and consumer satisfaction. This study explores the use of ultrasonic treatment (UT) to enhance the gel strength and water-holding capacity (WHC) of minced pork, and develops a multimodal, non-destructive prediction model based on fused spectral and image data. UT was applied at varying durations, with a 20-min treatment yielding optimal gel strength (338.2 g × cm) and WHC (77.87 %). To understand the physicochemical mechanisms behind these improvements, Raman spectroscopy and image-based texture analysis were employed. Raman results showed significant alterations in protein secondary structure, including unfolding and reorganization, while Gray Level Co-occurrence Matrix (GLCM) analysis of gel surfaces indicated increased structural uniformity and reduced randomness. These complementary features were integrated using a low-level data fusion strategy and modeled using Extreme Learning Machine (ELM), Support Vector Machine (SVM), and Convolutional Neural Network (CNN). The CNN model trained on augmented fused dataset achieved the highest prediction accuracy (Rp = 0.8954 for gel strength; Rp = 0.8887 for WHC), demonstrating the potential of combining chemical and spatial descriptors for real-time quality monitoring. This study not only confirms the effectiveness of ultrasound in improving pork gel quality but also introduces a robust and interpretable framework for intelligent meat processing and non-invasive quality assessment.