Plant leaf disease classification using deep attention residual network optimized by opposition-based symbiotic organisms search algorithm

• Published in: Neural computing & applications Vol. 34; no. 23; pp. 21049 - 21066
• Main Authors: Pandey, Akshay, Jain, Kamal
• Format: Journal Article
• Language: English
• Published: London Springer London 01.12.2022; Springer Nature B.V
• Subjects: Accuracy; Artificial Intelligence; Artificial neural networks; CAD/CAM; Color imagery; Computational Biology/Bioinformatics; Computational Science and Engineering; Computer Science; Data Mining and Knowledge Discovery; Deep learning; Economic impact; Feature extraction; Image classification; Image Processing and Computer Vision; Machine learning; Medical imaging; Momentum; Optimization; Original Article; Plant diseases; Probability and Statistics in Computer Science; Search algorithms; Sustainable development; Training; Plant leaf disease; Attention residual learning; Convolutional neural network; Opposition-based symbiotic organisms search (OSOS) algorithm
• ISSN: 0941-0643; 1433-3058
• DOI: 10.1007/s00521-022-07587-6

The main obstacle in front of the sustainable development of the agricultural sector is the considerable amount of economic loss due to reduced food production because of plant diseases. Computer-aided diagnosis of plant health conditions has paved its way in recent times by employing deep learning techniques especially convolutional neural networks (CNNs). The existing techniques mainly attained high classification accuracy if the images are captured in laboratory environments. Application on real world in-field images reduces their accuracy level significantly. To overcome the above shortcoming, this article merged the attention learning mechanism with the residual learning blocks and used the attention residual learning (ARL) mechanism for discriminative feature extraction from the RGB images of plant leaves. By employing the ARL strategy in the standard ResNet-50 CNN model, a new CNN module named AResNet-50 is designed for successful leaf disease recognition. Further, to reduce the chance of accuracy decrement due to erroneous choice of the training hyperparameters, Opposition-based Symbiotic Organisms Search (OSOS) algorithm is implemented for optimizing the values of learning rate and momentum during the training process. The efficacy of the proposed optimally tuned attention residual learning network, OSOS-AResNet-50, is checked on a leaf database created by the authors. Fifteen health conditions of citrus, guava, mango, and eggplant leaves are identified from their RGB images captured in real world or practical environment. The obtained classification accuracy is 98.20%. The experimental outcome reveals the superiority of OSOS-AResNet-50 over existing standard and largely used CNN models like AlexNet, VGG-16, VGG-19 and ResNet-50. Further, investigations disclose the importance of optimal training hyperparameter tuning and shows that approximately 2% more accuracy can be obtained by finding optimal values of learning rate and momentum with the help of OSOS.