Enhancing the severity classification of tomato plant epidemic pathogens using adaptive segmentation of mask RCNN and multiscale recurrent MobileNet

Tomatoes are the most significant and widely consumed crops globally. Leaf diseases cause an important threat to crop production and quality. Further, various fungi, bacteria and viruses can influence the plant's various parts and gradually destroy the quality and production of tomatoes in an agricultural field, thus it impacts the surrounding cultivated plants to cause more economical loss to farmers. Therefore, various techniques are proposed recently to optimally recognize and categorize the epidemic pathogens. To enhance sustainable plant protection practices, accurate identification and classification of diseases is essential to enhance the production rates. Effective pathogen detection and monitoring of plant health are critical areas of research in agriculture. Understanding disease severity is a crucial role for management practices and preventing the spread of infections. Rapid assessment is crucial because early detection of disease can significantly enhance the crop yield and influence the management strategies implemented by farmers. In this proposed model, a deep learning approach is proposed to classify the severity of diseases in tomato plants. At first, the needed images are collected from publicly available resource. Further, the collected images are subjected to the Adaptive and Attention-based Mask Region Convolutional Neural Network (AA-MRCNN) for optimally segmenting the abnormal regions from the gathered image. Further, the hyperparameters, like epoch, steps per epoch, and hidden neuron count in the Adaptive and Attention-based Mask Region Convolutional Neural Network are tuned by the Fitness-based African Vultures Optimization (FAVO) algorithm. Also, the segmented images are passed into Multiscale Recurrent MobileNet (MRMNet) module for categorizing disease severity in the tomato plant epidemic. The assessment of the recommended severity detection approach of tomato plant disease is determined by conducting a simulation experiment. The proposed model attains better outcomes of 93% accuracy, 93% specificity, 93% precision, 7% False Negative Rate (FNR), 86% Matthews Correlation Coefficient (MCC), 93% Fowlkes mallow Index (FM), 86% Bookmaker Informedness (BM), and 86% Threat Score (TS) measures in the ReLu activation function, which is progressed than the conventional frameworks. The result defines that the suggested technique outperformed than other baseline models to ensure the dependability of the tomato plant epidemic pathogens detection performance.