Milky WAI: Unlocking the Secrets of Raw Cow Milk Through Speckle Pattern and AI

This paper introduces a novel method for raw cow milk classification combining speckle pattern (SP) imaging and AI-based processing of statistical parameters. Raw cow milk has been classified considering (i) 6 statistical features extracted from raw SP images, and (ii) 4 features extracted from the Gray-Level Co-occurrence Matrix (GLCM) computed on each SP image. We conducted 4 experimental campaigns, resulting in 24, 000 SP images retrieved from 20 milks produced by 20 cows. We then considered two different datasets: (i) a Complete dataset (made of all 24,000 frames), and (ii) a Reduced dataset, built by excluding data from one acquisition campaign. The two datasets were then split into training and testing sub-datasets. We trained three Wide Neural Networks (WNN) using different strategies. WNN1 and WNN2 have been trained and tested using the Complete and the Reduced datasets respectively, while WNN3 has been trained on the Reduced tested on the Complete dataset. WNN3 only attained 72% accuracy (revealing sensitivity to environmental conditions), while WNN1 and WNN2 achieved a test accuracy higher than 90%, demonstrating effectiveness without relying on computationally expensive models such as CNNs. This approach could be considered as a promising initial step for cow milk classification using SP imaging.