Stock Photo: Getty Images
Food waste is a perplexing problem that weighs heavily on the global food production, production, distribution and sales industry, but a new generation of AI sensors is offering a host of fresh solutions.
The adoption of AI in the food industry is rapid, which is why Flinders University researchers have collaborated with an international research team to build the first comprehensive overview of AI technologies relevant to the food industry.
“There are many examples of AI-integrated sensors being used in the food industry to ensure product safety, maintain quality and optimize production efficiency,” explains Flinders University ARC Future Fellow Associate Professor Vi-Khanh Truong.
“This review highlights the strong potential of AI-integrated sensing systems to reduce energy consumption, fuel usage and food waste across the supply chain.”
As an example, AI-assisted precision drying systems can identify optimal processing conditions in real-time, significantly reducing excessive energy consumption during the food dehydration process. Similarly, smart spoilage prediction systems prevent premature food waste and reduce both economic losses and greenhouse gas emissions associated with food waste.
“The global food supply chain faces a growing need for monitoring systems that are not only accurate, but also rapid, non-destructive and scalable. This is because traditional laboratory methods for assessing food quality and safety, such as gas chromatography, microbial plating and sensory panels, are often destructive, time-consuming, create bottlenecks that prevent real-time quality control, and require specialized personnel.”
Researchers identified a wide range of intelligent sensing systems being integrated into the food industry, including AI-enabled optical sensors, hyperspectral imaging systems, electronic noses (e-noses), electronic tongues (e-tongues), Raman spectroscopy, FT-IR spectroscopy, microwave sensing platforms, IoT integrated low-power sensors, graphene chemosensors, and plasmonic sensors. ML-assisted multisensory arrays.
“By enabling real-time monitoring and predictive analysis, these intelligent systems can optimize food processing conditions, reduce unnecessary transportation and storage losses, minimize refrigeration energy demand, and improve logistics efficiency,” said Associate Professor Truong.
This review covered key AI frameworks such as support vector machines (SVMs), random forests, k-nearest neighbors (KNNs), convolutional neural networks (CNNs), long short-term memory (LSTM) models, autoencoders, and ensemble learning systems used for food quality, spoilage detection, adulteration analysis, and supply chain optimization.
Notable examples highlighted in the review include Raman spectroscopy combined with SVM to achieve up to 99.6% accuracy for adulteration detection in milk, FT-IR spectroscopy integrated with an AI model to achieve 100% classification accuracy for edible oil certification, and hyperspectral imaging using a CNN model to enable early detection of diseases in peppers before visible symptoms appear.
The integration of low-power IoT sensors with the TinyML edge computing platform is particularly important because these systems operate with minimal energy requirements and enable continuous monitoring directly within storage, transportation, and production environments.
“Taken together, these technologies support a more sustainable and resource-efficient food industry by reducing waste generation, reducing fuel and electricity consumption, and improving sustainability across the supply chain.”
The review also discusses AI-assisted electronic nose systems that can identify the geographic origin of coffee beans with 97.5% accuracy, as well as machine learning-assisted spoilage prediction systems for meat, fish, fruit, and dairy products.
These technologies demonstrate how AI can improve food safety, reduce waste, enhance traceability, and enable real-time quality monitoring across the supply chain.
“The choice of machine learning model primarily determines the regression error and prediction accuracy,” says Associate Professor Truong. “This study demonstrated the ability of machine learning models to improve sensor response across a variety of ambient conditions such as temperature, pH, humidity, and pressure.”
Wireless communication, used in conjunction with machine learning-assisted sensor arrays, also improves network efficiency, optimizes resource utilization, and improves predictive analytics.
Associate Professor Truong now predicts that the combination of machine learning models and sensors used in the food industry will rapidly increase, further improving target analysis selection and achieving full accuracy with fewer machine learning training cycles.
“While food sensors have been fabricated using a variety of nanomaterials due to their superior electrochemical properties, we believe that the selection of these sensors can be improved to detect more possible assays. This will allow sensing systems to consist of multiple machine learning models to detect a wide range of contaminants in food materials.”
