Plant disease identification using contextual mask auto-encoder optimized with dynamic differential annealed...

Most of the food consumed worldwide is produced by plants. Plant disease is a major cause of reduced production, but can be managed with regular monitoring. Manually observing plant diseases takes more time and is error-prone. Early detection of plant diseases with the aid of artificial intelligence and computer vision can decrease the effects of disease and help plants withstand the downsides of continuing surveillance. In this manuscript, plant disease identification using contextual mask auto-encoder optimized with dynamic differential annealed optimization algorithm (PDI-CMAE-DDAOA) is proposed. The plant village dataset is used to collect the images. Then the image is fed to preprocessing. Using an adaptive self-guided filter approach, the noise is removed from the input images during the pre-processing phase. The result of the pre-processing section serves as input for the feature extraction segment. Four statistical features, including mean, variance, entropy, and kurtosis, are recovered from the cosine similarity hidden Markov model (CSHMM). The contextual mask auto-encoder (CMAE) is given the extracted features to accurately classify the healthy and unhealthy regions of the plant image. The issue of slow convergence affects the CMAE. However, it is noted that the CMAE converges more quickly with deep learning features than with texture features in this instance. The CMAE classifier generally does not exhibit any adaptation of optimization algorithms for determining the best parameters to ensure the precise classification of plant disease. Therefore, dynamic differential annealed optimization algorithm (DDAOA) is considered to enhance the CMAE classifier, which accurately distinguishes between healthy and diseased plants. The proposed PDI-CMAE-DDAOA is done in Python. The efficacy of PDI-CMAE-DDAOA is evaluated under some performance metrics, like accuracy, precision, sensitivity, F1-score, specificity, error rate, receiver operating characteristic curve (ROC), computational time. The proposed method provides higher accuracy 23.34%, 34.33%, and 32.07%; higher sensitivity 36.67%, 36.33%, and 23.21%; higher F1-score 46.67%, 57.56%, and 43.21%; higher specificity 56.67%, 67.56%, and 23.21% analyzed with existing models, like transfer learning-based deep ensemble neural network for plant leaf infection recognition (PDI-DENN), plant disease detection with hybrid model based on convolutional auto-encoder and convolutional neural network (PDI-CAE-CNN), and automatic and reliable leaf disease finding depending on deep learning methods (PDI-EN-CNN), respectively. RESEARCH HIGHLIGHTS: To find the plant disease at early stage. To present PDI-CMAE-DDAOA. To get better classification accuracy by extracting the optimal features with the help of efficient CSHMM. To minimize the error during classification process. To maximize high area under curve value.