Generative Adversarial Networks for Precision Animal Management: A Review

Highlights The GANs have been applied on the data modalities of RGB image, wearable sensor data, ultrasound image, thermal image, and hyperspectral image in animal farming. The GANs can increase accuracy by over 14% in animal management via augmenting development datasets. The GANs could generate unrealistic images, and more advanced model architectures and training strategies are needed. ABSTRACT. In modern artificial intelligence (AI) modeling, achieving optimal classification and/or regression performance is keenly pursued when applying these models to improve animal production systems. Large-scale, balanced, and high-quality datasets are tremendously beneficial for developing robust and generalizable AI models that learn data representations directly from the training process. Livestock and poultry are complex, individually different, time-varying, and dynamic living organisms, creating significant challenges due to biological variability and unstructured data collection environments. Limited laboratory capacities and restrictions on animal usage also hinder the creation of large datasets. Data augmentation can automatically generate synthetic data to expand datasets, boosting model performance while reducing manual efforts in data collection and annotations. Besides classical image processing-based data augmentation techniques, generative adversarial networks (GANs), consisting of two competing neural networks (a ‘generator’ and a ‘discriminator’), provide a novel approach that can learn to generate realistic data samples by having the networks iteratively challenge each other. There is an increasing awareness of utilizing GANs and their variants for data augmentation and synthesis in precision agriculture to improve AI model performance. This review presents an overview of the evolution of GAN architectures along with the applications in animal farming domains, covering a wide range of animal species, farming systems, and production phases. Applications as well as the challenges and opportunities of GANs are discussed for future research. Keywords: Animal management, Artificial intelligence, Data augmentation, Deep learning, Precision agriculture.