Thermal Imaging-Based Subclinical Mastitis Detection with ANOVA-Enhanced Colour Feature Selection Using Regression Model

Efficient detection of subclinical mastitis in dairy cattle is essential for preventing long-term health issues and ensuring optimal milk yield. The proposed subclinical mastitis detection system uses colour feature extracted from infrared thermographic images of cow udders. Colour features are extracted from three prominent colour spaces—RGB, LAB, and HSV—wherein statistical descriptors such as mean, standard deviation, variance, skewness, and kurtosis are calculated for each channel to capture key image properties. To optimize feature selection, statistical techniques such as ANDV A is applied to filter the most relevant and high-ranking features, reducing dimensionality and computational load. A custom dataset, consisting of infrared thermographic images captured from the dairy farms in the southern region of Tamil N adu under diverse thermal imaging modalities, including Arctic, Lava, and others, was developed to reflect varying thermal conditions. Various machine learning models, such as Support Vector Machines (SVM), Random Forest, and Gradient Boosting, are trained and evaluated based on their performance using metrics like accuracy, and cost. The findings reveal the use of kernel logistic regression obtained accuracy of 94.5% and cost effective for the dataset.