MULTIPLE PLANT LEAF DISEASE CLASSIFICATION USING DENSENET-121 ARCHITECTURE
Journal: International Journal of Electrical Engineering and Technology (IJEET) (Vol.12, No. 5)Publication Date: 2021-05-31
Authors : Aswin Vellaichamy S Akshay Swaminathan C Varun Kalaivani S;
Page : 38-57
Keywords : metallic surface; VGG-16; Neural Networks; defect detection;
Abstract
Agriculture is the backbone of a country in terms of economy and survival of the people. To maintain a high efficiency of crop production we look to avoid plant diseases. The proposed algorithm is to optimize the information from the resources available to us for the betterment of the result without any complexity. The neural network used for classification is the Dense Convolution Neural Network (DCNN). In this project, a pre-trained neural network model (densenet-121) which isimported from the keras library has been used for training. Aconvolution may be a simple application of a filter to an input that leads to activation. Frequent application of equivalent filterto an input, leads to a map of activations called feature map, indicating the locations and strength of a detected feature in an input such as an image. The convolutional networks help to automatically learn an outsized number of filters in parallel specific to a training dataset. This algorithm helps provide an efficient result in detecting plant diseases, which in turn helps the economy of the country as well. Using 35779 images from Huges DP Plant- Village dataset from Kaggle, the densenet-121 has been used to classify the 29 different diseases for 7 plants (potato, tomato, corn, bell pepper, grape, apple and cherry). In this project, the original image is converted to HSV color form and then the masked image is generated by thresholding and given to the proposed model for training and classification, giving an average accuracy (theoretical) of 98.23%. When all classes of plant disease are given together to the model for training on google colab platform, (Tesla-T4 processor) we got an average accuracy of 94.96% for 50 epochs with a learning rate of 0.002. Additionally, a basic user-friendly website to test the trained model for the disease affected plant images and get prescription for the plant disease. Further the scope to extend the dataset used for training to identify many plant diseases
Other Latest Articles
- DEEP METALLIC SURFACE DEFECT DETECTION
- ROBUST DELAY-DEPENDENT STABILITY FOR TAKAGI-SUGENO FUZZY SYSTEMS WITH MIXED TIME-VARYING DELAYS
- ADAPTING DIGITAL BANKING SERVICES DURING COVID-19 PANDEMIC: A STUDY ON CUSTOMER EXPERIENCES IN SELECTED DISTRICT OF MAHARASHTRA STATE, INDIA
- Modélisation De La Demande En Eau Dans Une Région Aride. Cas De La Wilaya De Djelfa
- AN EFFECTIVE MECHANISM FOR SECURING AND MANAGING PASSWORD USING AES-256 ENCRYPTION & PBKDF2
Last modified: 2021-06-04 21:13:14