Deep Learning Ensemble Approach for Classification of Indigenous Pig Breeds in India

Abstract

Breed identification plays a crucial role in conserving indigenous pig breeds, managing genetic resources and developing effective breeding strategies. Traditional methods of breed recognition are often time-consuming, subjective and prone to errors. Therefore, an automated mechanism is required to replicate expert skills for pig breed identification within farm environment, providing significant benefits to the farmers in the piggery sector. To fulfil this necessity, we propose ensemble learning-based approach for the automated identification of indigenous pig breeds in India using models based on convolutional neural networks (CNNs). Four transfer learning models such as EfficientNetB0, Xception, InceptionV3 and InceptionResNetV2 models are combine to construct stacking ensemble model. In this approach, multiple meta-learners including SVC (Linear), Logistic Regression, Ridge Classifier, SGD Classifier and MLP Classifier, were used to enhance the ensemble performance and the best-performing combination was selected. For this study, we used a dataset of 2040 images with eight distinct indigenous pig breeds of India. Experimental results demonstrate that the proposed stacking ensemble approach with Ridge Classifier achieves superior classification accuracy of 98% outperforming individual models by 3.5–8.3%, highlighting its potential as a practical tool for smart pig farming, breed conservation and genetic resource management. The study emphasizes the application of vision intelligence and ensemble learning in precisely and efficiently distinguishing between various indigenous pig breed and management in India