Poultry diseases pose a significant threat to the global poultry industry, impacting both economic profitability and public health. Early and accurate diagnosis is crucial for effective treatment and prevention of these diseases. This study investigates the potential of deep learning for automated classification of four common poultry diseases: Coccidiosis, Healthy, Newcastle Disease, and Salmonella, using fecal images. We propose a novel ensemble model, combining two independently trained EfficientNetB3 and MobileNet architectures, to enhance classification accuracy. Our methodology incorporates a comprehensive preprocessing pipeline, including image resizing, ICA whitening, and image standardization, to optimize the input data for the model. To address the challenge of limited data availability, we employ a data augmentation strategy, effectively expanding the training dataset. The research utilizes a dataset of 8067 annotated poultry fecal images, providing a diverse representation of different disease conditions. The proposed approach achieves a remarkable accuracy of 98.33% in classifying poultry diseases, surpassing the performance of individual CNN models and recent studies in this field.