Menu

Menu controller icon

Analysing milk quality in real time using edge intelligence

Posted: 04-10-2024

Tags:

    This paper introduces the AppAdapt-LWDL framework, a novel approach to enable real-time milk species identification and adulteration detection on resource-constrained edge devices.

    Outline the paper

    This paper, funded by VistaMilk SFI Research Centre, investigates a solution to enhancing milk quality testing in dairy farms and processing plants on-site. Instead of sending large amounts of data to a cloud server hosted elsewhere, the paper introduces a new method to process this data right where it is collected — at the “edge” of the network. To do this, they use a special type of artificial intelligence (AI) that has been simplified to work efficiently on smaller devices like those found on farms. The AI models are made smaller by removing unnecessary parts and simplifying how they work, making them faster more energy efficient without losing accuracy. This allows the quick detection of what type of milk is being processed and whether it has been mixed with other types ensuring quality and safety in real-time.

    What does it mean to me?

    Samples of dairy products being tested

    This work has significant real-world implications, particularly for improving food safety and transparency in dairy products. By allowing real-time identification of milk type and detecting adulteration directly at dairy farms or processing facilities, it ensures that the milk reaching consumers is pure, safe, and properly labelled. For people who are allergic to certain types of milk or who prefer specific varieties (like goat or cow milk), this technology can ensure they receive exactly what they expect, reducing health risks and dietary concerns. It can also help producers comply with regulations and reduce fraud by preventing the mixing of different milk types for profit. Additionally, the efficient use of AI on small devices means these quality checks are faster, cheaper, and use less energy, which can help lower costs and make milk quality monitoring accessible even in remote or rural areas.

    What does it mean for the dairy industry?

    The AppAdapt-LWDL framework has promising industry applications for real-time quality monitoring in dairy processing. Currently, it can be used in dairy farms, milk collection centres, and processing plants to rapidly identify milk types and detect adulteration, ensuring product authenticity and compliance with regulations. This helps producers maintain quality control, reduce fraud, and improve consumer trust. In the future, the framework’s ability to be deployed on edge devices means it can be embedded into handheld sensors or equipment used by quality inspectors, making the technology accessible in remote or rural areas without reliance on cloud computing. Additionally, it could be integrated into automated production lines for continuous monitoring, reducing the need for manual checks. Beyond the dairy sector, this technology has potential applications in other industries, such as meat processing, beverage quality control, and pharmaceuticals, where real-time and on-site analysis of product quality is essential.

    What is the future of this research?

    The future steps for this research involve enhancing the AppAdapt-LWDL framework’s capabilities and expanding its applications within the dairy industry and beyond. One potential direction is to improve the accuracy and efficiency of the model further by incorporating more advanced AI techniques or utilizing more diverse datasets, which can adapt to different milk compositions and regional varieties. The framework could also be extended to monitor other quality aspects of dairy products, such as fat content, bacterial contamination, or shelf-life prediction. Beyond dairy, the same approach could be applied to other areas of food processing where real-time quality assessment is crucial, such as meat processing, fruit grading, or packaged food safety. Furthermore, integrating this technology with other Internet of Things (IoT) systems could facilitate automated supply chain monitoring, ensuring quality from farm to consumer in a wider variety of agricultural and food industries.

    Publication Title: Application Adaptive Light-Weight Deep Learning (AppAdapt-LWDL) Framework for Enabling Edge Intelligence in Dairy Processing

    Authors: Rahul Umesh MhapsekarSteven DavyLizy Abraham, Indrakshi Dey

    Publication Date: 1st October 2024

    Name of Journal: IEEE Transactions on Mobile Computing