Designing a New Tomato Leaf Disease Classification Framework using RAN-based Adaptive Fuzzy C-Means with Heuristic Algorithm Model

Abstract

In tomato production, one of the most significant problems is the identification of Tomato Leaf Disease (TLD). Plant leaf disease is the primary factor that influences both the quality and quantity of crop production. India holds the second position in tomato making. However, multiple diseases contribute to the decline in the quality of tomatoes and the decrease in crop yield. Hence, it is important to accurately categorize and diagnose the tomato plant leaf infection. The productions of tomatoes are impacted by many leaf diseases. Early recognition of the diseases helps to reduce the disease infection and improve the yield of crops. Certain diseases are identified and шlassified using several methods. Therefore, the TLD classification and identification model is developed to solve the above problems. The images related to tomato leaves are aggregated in the initial phase through online sources. Then, the images are forwarded to the pre-processing phase. Further, the pre-processed image is given to the segmentation process, where the Adaptive Fuzzy C-Means (AFCM) technique is utilized. Meanwhile, the parameters of the AFCM algorithm complicate the cluster assignment in the presence of outliers or noise, thus resulting in reduced clustering performance. So, the parameters of AFCM are tuned by utilizing the new improved algorithm named Dingo Optimization Algorithm (DOA) to improve the clustering accuracy. It is done by assuming the AFCM parameters as a population of Dingoes and the maximum classification accuracy as its fitness function. Finally, the segmented images are fed to the classification process, where the Residual Attention Network (RAN) is used to attain the classified outcomes. Therefore, the investigated system shows a more efficient TLD prediction rate compared to traditional techniques in the experimental investigation. The results from the experiments indicate that the suggested models exhibit exceptional classification performance, achieving an accuracy rate of 95.22%. Therefore, the model suggests advancement in predictive capabilities over traditional methods.