![]() | Only 14 pages are availabe for public view |
Abstract Agricultural development is crucial for addressing the challenge of feeding a growing global population, projected to reach nearly 10 billion by 2050. Plant diseases pose a significant threat to food production, impacting both quantity and diversity. Early detection of these diseases is essential for improving food production quality and reducing economic losses. Existing models for automatic plant disease detection, while accurate for single plant types, struggle with low-resolution images and complex backgrounds, limiting their effectiveness in real-world scenarios. Prior methods rely on high-quality images for accurate classification, rendering them ineffective with low-resolution images, crowded backgrounds, shadows, and varying textures and brightness. Additionally, while models like MobileNet and EfficientNet offer compact sizes and parameter efficiency, they may not capture essential features adequately for accurate disease diagnosis when used individually. Most previous models focus on detecting diseases in a single plant type rather than across multiple plants and various plant diseases. To address these limitations, proposes an innovative approach combining MobileNet and EfficientNet, leveraging their complementary strengths to enhance plant disease detection performance. By utilizing a pre-trained model based on convolutional neural networks (CNNs), we aim to achieve higher accuracy even in challenging conditions such as lower resolution images and diverse backgrounds. Our strategy involves exposing the neural networks to deliberately noisy training datasets, optimizing them to generalize effectively and perform robustly in real-world scenarios. |