Integrating NMSA based advanced light-weight aggregated fusion channel network for robust tomato leaf disease detection

Important agricultural sectors in our nation is tomato production, which is vital for both domestic consumption and international trade. The early finding besides control of diseases affecting tomato leaves is crucial since these diseases directly impact tomato yield and quality. Tomato yields are increased and farmers are able to better handle disease management when automated disease detection in tomato leaves is implemented. Encouraging optimal tomato plant growth and ensuring a plentiful supply to fulfil the growing worldwide demand and offer food security hinges on the prompt detection, control, and resolution of tomato leaf specificity. Presently, there is extensive use of computer-assisted technology for the finding and diagnosis of plant diseases. The project builds the lightweight aggregated fusion channel network (LAFCN) to aid in the investigation and detection of diseases. Using photos from a standard dataset, an optimised capsule neural network (CapsNet) is trained to identify and categorise ten different tomato leaf diseases. Data augmentation and preprocessing approaches were used during training to reduce the likelihood of overfitting. Because of its better capacity to capture spatial placement inside the image, CapsNet was selected over CNNs. To increase classification accuracy, Non-Monopolize Search Algorithm (NMSA) appropriately selects relevant features. We also provide a better localization loss tailored to tiny-affected region prediction and incorporate a Contextual Transformer (CoT) block to boost localization accuracy. The results show that the enhanced ELA-C3 module achieves better feature extraction and detection accuracy in experiments, and that the suggested WGC-PANet achieves a lightweight state by plummeting the computational complexity and quantity of model parameters. Furthermore, the parameters of the suggested prototypical are fine-tuned using the Hunger Game Search Algorithm (HGSA). Various measures, including \(\:accuracy,\:precision,\:recall,\:besides\:F1-score\) $accuracy,precision,recall,besidesF1-score$ , are used in the studies, which are conducted on publically available datasets. The results demonstrate that, in judgement to existing models, the suggested model attained an accuracy, precision, recall, and F-score ranging from almost 98–99%.