Plant diseases are a major threat to agricultural productivity worldwide, making early and precise identification essential for efficient disease management. Traditional methods of identifying crop diseases are time-consuming, and often subjective. This study addresses the challenge of developing a generalizable crop disease classification model capable of handling diverse plant species and disease types to facilitate timely disease mitigation. Using a dataset compiled from multiple open sources, containing images captured in both real-world field settings and controlled lab environments, the study leverages transfer learning with fine-tuning to improve model performance. The dataset includes 39 healthy and disease classes across 17 crop species, providing a robust basis for multi-species disease identification. The methodology involved training several pre-trained convolutional neural network models, including VGG19, DenseNet201, InceptionResNetV2, MobileNetV2, and ResNet152V2. Models were initially trained with frozen layers, followed by fine-tuning through selective layer unfreezing to enhance their ability to recognize disease features in the highly imbalanced dataset. After fine-tuning, VGG19 model achieved 98.15% test accuracy, 0.9996 AUC-ROC value, 0.9701 F1 score, 0.9916 precision, and 0.9739 recall value, demonstrating superior performance in distinguishing between a wide range of crop diseases. These results highlight the effectiveness of transfer learning in multi-species plant disease classification, predominantly for datasets with complex and varied image sources. This research provides a generalizable, scalable, high-performance model for automated disease detection, supporting sustainable agriculture and enhancing agricultural practices through timely and accurate disease management.