1. Plant leaf disease classification using deep Convolutional neural network with Bayesian learning
- Author
-
Preeti Singh, Guneet Sachdeva, and Pardeep Kaur
- Subjects
010302 applied physics ,education.field_of_study ,Computer science ,business.industry ,Bayesian probability ,Population ,02 engineering and technology ,Overfitting ,021001 nanoscience & nanotechnology ,Residual ,Bayesian inference ,Machine learning ,computer.software_genre ,01 natural sciences ,Convolutional neural network ,Data set ,0103 physical sciences ,Artificial intelligence ,0210 nano-technology ,business ,education ,computer ,Feature learning - Abstract
In the present situation, the world's population continues to rise rapidly, while the cultivated land suitable for cultivation stays the same. It forces farmers to demonstrate innovative techniques that improve crop yields to sustain the increasing population. Preserving the health of crops is quite critical in this respect. Therefore, early detection of disease in crops seems to be very crucial to monitor disease and increase crop yield. In this paper, a deep Convolutional Neural Network (DCNN) model is proposed using Bayesian learning process for classification of the diseases in plants. Material used for this work consist of 20,639 images taken from plantVillage having 15 distinct classes of healthy and infected leaf images of potato, tomato and pepper bell. The key objective of this research is the introduction of Bayesian process at the top of residual network for efficient feature learning. An investigation was performed for comparison of proposed DCNN with the traditional classifiers in terms of accuracy, precision, recall and F-score and result indicated that the proposed model is an efficient tool for classification of diseases. Further, the prominent physical or chemical properties of nanomaterials have enabled their use as innovative and high-performance diagnostic tools for varied plant pathogens and other important disease biomarkers. Taking these as a pre processing step while acquiring the data set and combining it with proposed work enhances the performance. The proposed approach utilizes the features developed by CNN mostly in different hierarchies to use Bayesian framework and obtains the accuracy rate of 98.9% with no sign of overfitting.
- Published
- 2021
- Full Text
- View/download PDF