Fuzzy-Filtered Neural Network for Rice Disease Diagnosis using Image Analysis
Toran Verma1, Sipi Dubey2
1Toran Verma, Department of Computer Science & Engineering, Rungta College of Engineering and Technology, Bhilai (C.G.), India.
2Sipi Dubey, Department of Computer Science & Engineering Rungta College of Engineering and Technology, Bhilai (C.G.), India.
Manuscript received on 09 June 2019 | Revised Manuscript received on 14 June 2019 | Manuscript Published on 08 July 2019 | PP: 437-446 | Volume-8 Issue-8S3 June 2019 | Retrieval Number: H11050688S319/19©BEIESP
Open Access | Editorial and Publishing Policies | Cite | Mendeley | Indexing and Abstracting
© The Authors. Blue Eyes Intelligence Engineering and Sciences Publication (BEIESP). This is an open-access article under the CC-BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)
Abstract: Image mining plays a vital role in the decision-making process in many application areas. Image mining is part of information processing and management. Plant diseases compromise productivity which impacts social life and economy of the nation. The effective use of agriculture image mining can enhance yield production and give economic benefit to the farmer and the country. The research aimed to automate rice diseases identification using image mining for quick diagnosis of the diseases. The digitally captured disease infected and disinfected plant images stored in the database which carries unique feature descriptor in the form of color information, texture appearance, and spatial-frequency information. In this research, the digitally acquired five categories of infected and one category of disinfected images stored in JPEG format in the database. Each category defines unique image features. The acquired images are accessed in RGB color space and cropped and resized in pre-processing steps. All pre-processed images are segmented using Otsu’s two-level threshold on a* components of L*a*b* color space image. The segmentation process generates three segments for each image. The 54 hybrid features are extracted using image analysis which includes 6 color entropy, 24 texture, and 24 wavelets F-ratio of spatial-frequency components. The two-way ANOVA analysis is applied in wavelet features to evaluate F-ratio. The extracted features are passed in the CART to select relevant features according to the Gini index split point. The CART created a binary decision tree, reduces 54 attributes to only 13 relevant attributes. The CART selected 13 attributes forwarded in FIS for fuzzy filtering which summarized 13 attributes to 6 attributes. The fuzzy filtered outcomes used to train MLPNN using Scaled Conjugate back-propagation training algorithm to design rice diseases recognition model. The CART feature selection and fuzzy filtering process applied to summarize relevant input features which reduce the complexity of MLPNN. The hybrid CART-FIS-MLPNN model gives 97.1% training and 95.47% testing efficiency.
Keywords: Two-way ANOVA, Classification and Regression Tree (CART), Fuzzy Inference System (FIS), Multilayer Perceptron Neural Network (MLPNN), Image Processing, Pattern Recognition.
Scope of the Article: Image Processing and Pattern Recognition