A Deep Transfer Learning Approach for Early Identification of Multiple Crop Diseases and Pest Infestations using Leaf and Field Images
Abstract
Effective and early detection of plant diseases and pest infestations is vital for increasing agricultural efficiency to provide global food security. Nevertheless, traditional diagnosis methods based on manual visual inspections are inefficient, inconsistent, and inaccurate in practical scenarios. In this research, we propose the application of deep transfer learning technique to detect multiple plant diseases and pest infestations using agricultural images acquired from heterogeneous crop level environments and leaf level environments. Specifically, five pre-trained CNN architectures, i.e., VGG16, ResNet50, DenseNet121, MobileNetV2, and EfficientNetB0 are applied within a transfer learning setting for ImageNet data set to extract and classify features efficiently. A series of image pre-processing operations, such as resizing, normalization, noise reduction, and contrast enhancement, are used to enhance the quality of images while various data augmentation techniques help improve model generalizability across different environments. In particular, the transfer learning model is trained on a mixed data set including 20,000 heterogeneous images of healthy plants, bacterial, fungal, and viral plant diseases as well as pest infestations. According to our experimental evaluations, DenseNet121 achieves the best accuracy rate at 98.21%, while EfficientNetB0 is second only in classification task. However, MobileNetV2 exhibits high accuracy and faster inference speed. The comparative analysis confirms that transfer learning significantly enhances robustness, generalization capability, and computational efficiency under real-world farming conditions. By integrating disease and pest detection within a unified framework, the proposed approach addresses key limitations of existing models that rely heavily on controlled datasets. Overall, this study contributes to precision agriculture by providing a scalable, accurate, and deployment-oriented solution for intelligent crop health monitoring systems.