A Novel Deep Learning Framework Approach for Sugarcane Disease Detection

• Published in: SN computer science Vol. 1; no. 2; p. 87
• Main Authors: Srivastava, Sakshi, Kumar, Prince, Mohd, Noor, Singh, Anuj, Gill, Fateh Singh
• Format: Journal Article
• Language: English
• Published: Singapore Springer Nature Singapore 01.03.2020; Springer Nature B.V
• Subjects: Accuracy; Advances in Computational Intelligence; Algorithms; Artificial intelligence; Bagasse; Classification; Classifiers; Computer Imaging; Computer Science; Computer Systems Organization and Communication Networks; Crop diseases; Data Structures and Information Theory; Datasets; Deep learning; Feature extraction; Geometry; Information Systems and Communication Service; Machine learning; Neural networks; Original Research; Paradigms and Applications; Pattern Recognition and Graphics; Regression analysis; Software Engineering/Programming and Operating Systems; Statistical analysis; Sugar; Sugarcane; Support vector machines; Vision; India; Deep learning; VGG-19; Inception v3; Sugarcane; VGG-16
• ISSN: 2662-995X; 2661-8907
• DOI: 10.1007/s42979-020-0094-9

Sugarcane, belonging to the grass family Poaceae, is rich in sugar sucrose, thereby used for making white sugar, jaggery and other by-products like molasses and bagasse. However, a diseased sugarcane plant is of no use, so it needs to be detected as soon as possible. A novel deep learning framework approach is proposed in this paper to detect whether a sugarcane plant is diseased or not by analyzing its leaves, stem, color, etc. The study comprises three scenarios based on different feature extractors namely Inception v3, VGG-16 and VGG-19. These are the pertained models on which different classifiers are trained. The state-of-the-art algorithms (SVM, SGD, ANN, naive Bayes, KNN and logistic regression) are compared with deep learning algorithms like neural network and hybrid AdaBoost. Several statistical measures such as accuracy, precision, specificity, AUC and sensitivity are calculated using Orange software, and the scenario having the highest accuracy is chosen. The receiver operating characteristic curve is computed in order to assess accuracy. An AUC of 90.2% is obtained using VGG-16 as the feature extractor and SVM as the classifier.