AI-Driven Framework for Multi-Modal Contaminant Detection in Meat Products Using CNN-LSTM and Sensor Fusion

The safety and quality of both poultry and meat products should be guaranteed by effective contaminant identification that can be done in real-time using low-power IoT devices. Common inspection procedures tend to be slow, labor intensive and inaccurate, especially in dynamic based industries. The challenge is met in this study, which proposes an AI-driven hybrid framework with additional spatial, temporal, and noise-reduction, one that improves the accuracy of detection and deployment feasibility. The new framework will combine Convolutional Neural Networks (CNN) to extract spatial features, Long Short-Term Memory (LSTM) networks to model sequences in time and a Kalman Filter to refine the created predictions and suppress noise. The model was tested and trained over a large variety of material by using the samples of contaminants and non-contaminants, and the feature optimization was obtained with the help of systematic experimentation. An ablation study was also used to determine the impact of individual components, and latency resource usage was tested on a Raspberry Pi 4 to confirm real-time implementation viability. The entire model attained a class prediction average of 96.5 percent and an IoU of 92.4 percent exceeding the current machine learning-based approaches to contaminant detection. Ablation findings showed high accuracy cost due to cancellation of whole or part of the CNN/LSTM, which indicates that those are critical components. Latency check revealed that edge devices perform an average of 0.42 seconds on inference per sample, with moderate CPU (68%) and memory (1.9 GB) utilization on edge devices showing operational efficiency. The results confirm the applicability of the hypothetical framework to provide precise, fast, and implementable food processing plant contaminant detection. Further research will center on extending diversity to the datasets, introducing the Explainable AI concepts to enhance the transparency, and modifying the framework to easily allow multi-class detection of the contaminants to develop an industry-ready intelligent quality control system that can scale.