Adaptive leaf abnormality segmentation and improved plant disease classification by tuned long short‐term memory with recurrent neural network

Abstract

The conventional models suffer in providing significant accuracy and detection speed, especially in detecting apple diseases, to assure the healthy development of the apple industry. The main aspect of the proposed scheme is to present a new plant disease classification using intelligent segmentation and classification models. The adaptive leaf abnormality segmentation is enhanced by the solution index-based Jaya-Krill Herd optimization (SI-JKHO). Further, the feature extraction is accomplished using "gray level co-occurrence matrix, hybrid local binary pattern with local gradient patterns," color features, and shape features. These collected features are subjected to feature selection using principle component analysis, in which the optimal features are obtained that are further considered to "tuned long short-term memory with recurrent neural network (T-LSRNN)." From the validating results, the performance of the enhanced-JKHO-LSRNN method is correspondingly secured at 6.66%, 6.66%, 7.86%, and 4.34% higher enriched performance RNN, long short-term memory, convolutional neural networks, and LSRNN at 35th learning percentage in dataset 3. The results confirm that the developed model can accurately determine plant diseases compared to conventional models based on diverse performance metrics like "accuracy, precision, specificity, sensitivity, false positive rate, and false negative rate," and so forth. K E Y W O R D Sadaptive fuzzy C-means clustering with thresholding, plant disease classification, solution index-based Jaya-Krill Herd optimization, tuned long short-term memory with recurrent neural network INTRODUCTIONPlant disease can be diagnosed through the plant leaves, which involve more complexities while considering optical observation. 1 Due to the increasing cultivation of plants, the complexities also increase along with the conventional phytopathological problems. Although experienced agronomists and pathologists diagnose plant diseases, sometimes it causes failures in diagnosing particular plant diseases and leads to wrong conclusions and further mistaken treatments. 2 The existing automatic computational system has offered significant support to the agronomist in detecting and diagnosing plant diseases while subjecting the optical observation of infected plants through their leaves. 3 Here, the developed system has to be simple