Deep CNN model for crops’ diseases detection using leaf images

• Published in: Multidimensional systems and signal processing Vol. 33; no. 3; pp. 981 - 1000
• Main Authors: Kurmi, Yashwant, Saxena, Prankur, Kirar, Bhupendra Singh, Gangwar, Suchi, Chaurasia, Vijayshri, Goel, Aditya
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.09.2022; Springer Nature B.V
• Subjects: Artificial Intelligence; Artificial neural networks; Circuits and Systems; Color imagery; Crop production; Datasets; Electrical Engineering; Engineering; Feature extraction; Image classification; Machine learning; Medical imaging; Neural networks; Signal,Image and Speech Processing; Leaf image preprocessing; CNN model; Identification and classification; Computer-aided diagnosis
• ISSN: 0923-6082; 1573-0824
• DOI: 10.1007/s11045-022-00820-4

The agricultural yield of any country provides the base for the development of that nation. Sustainable growth needs to maintain crop production up to a certain level that depends on the research of their disease detection and treatment. The general approaches available in the literature follow attributes extraction and training a classifier model for leaf image classification that limits accuracy. The proffered technique eliminates the redundant information from the image dataset. We initially localize the region of interest in terms of the color attributes of leaf image based on the mixture model for region growing. The feature extraction is performed through a proposed deep convolutional neural network model followed by the classification of the leaf images. The deep learning model uses color images to learn the attributes that show different patterns that can be distinguished with the help of a convolutional neural network model. The execution measure of the proposed model is investigated using the PlantVillage dataset. The simulating replica outcomes show that the performance of the proposed model is far better as compared to the existing well-known methods of the domain with mean classifying accuracy and area under the characteristics curve of 95.35% and 94.7%, individually.