Implementation of adaptive multiscale dilated convolution-based ResNet model with complex background removal for tomato leaf disease classification framework

The automatic identification of tomato leaf disease has been regarded as a subjective, laborite as well as time-consuming technique. It is crucial to identify the small discriminative features among various tomato leaf diseases. In addition to that, it has brought some difficulties to the fine-grained visual classification of tomato leaves-dependent images. Hence, it is necessary to develop a tomato leaf disease classification framework with an effective background removal technique. The newly proposed model has been effectively utilized to classify the tomato leaf disease and the complex background removal. The proposed model has removed the background without degrading the information preserved in the images. Thus, the proposed model has improved the accuracy rate. The tomato plant disease-based images are collected from the real-time dataset. Then, the collected tomato images are pre-processed using the contrast-limited adaptive histogram equalization and median filtering approach. It is then inserted into the data augmentation stage for increasing the data without collecting new data, where the super-resolution generative adversarial network is used. Further, the Deeplabv3 model is used for removing the background from the augmented images, which reduces the unnecessary portion of the images. The background removed images are utilized for the pattern extraction phase using an improved local gradient pattern, where the hybrid optimization algorithm of elephant herding spider monkey optimization (EHSMO) is developed for tuning the parameters in LGP to increase the classification performance. These extracted patterns are incorporated for tomato leaf disease classification, which is done by adaptive multiscale dilated convolution-based ResNet along with the EHSMO algorithm for parameter optimization. Finally, the severity computation is done to evaluate the severity level among classified outcomes.