SqueezeNet-Based Deep Learning Framework for Accurate Tomato (Solanum lycopersicum) Leaf Disease Diagnosis and Classification

Abstract Agriculture is crucial for food security but is severely threatened by crop disease and climate variability and causes severe yield loss. As the population grows worldwide, quick and accurate disease detection is critical. Deep learning, in particular through transfer learning, offers promising solutions, but most are computationally costly and unsuitable for real-time use in low-resource settings. There is limited research on lightweight models like SqueezeNet with optimized training parameters. This suggests the need for an efficient, high-accuracy, and deployable model to facilitate timely detection of tomato leaf diseases under real-world agricultural settings. This study presents a deep learning model based on the SqueezeNet framework for the detection and classification of tomato leaf diseases. Various combinations of optimizers (SGDM, ADAM, RMSProp) and learning rates (0.0004, 0.004) were employed during both training and testing phases, resulting in six configurations per case. The SqueezeNet model achieved 99.91% and 99.86% accuracy for TMC class classification during testing and training, with ADAM learning at 0.0004. ADAM at 0.0004 had ideal recall (100%) for the TH class during testing, and SGDM had 99.65% recall for the TYLCV class at the same learning rate, proving the model’s usefulness. The proposed framework is robust, with F1-Scores of 99.42% in ADAM testing at 0.0004 and 99.38% in SGDM training at 0.0004 for the TYLCV class. The model’s low misclassification rate (0–0.21%) boosts confidence. The ability to demonstrate classification performance and the minimal computational requirements of the proposed SqueezeNet-based system enhance the latter’s feasibility for use in real-time agricultural environments that are resource-constrained. Its scalability and resilience make it an excellent choice for utilization in advanced disease monitoring systems for tomato leaf diseases, facilitating quick, accurate diagnosis at the field level to facilitate enhanced precision agriculture practices.