Design and performance evaluation of paddy leaf disease identification using KNN algorithm and Keras model

This research aims to establish paddy leaf disease identification system by using Python programming language. Paddy rice is one of the important food crops in the world. Although paddy rice can be harvested twice per annum, the productivity could be affected if the paddyplantation is infested by diseases. Paddy leaf disease is one of the diseases that can cause huge loss to paddy crops and bring down the farmers’ income. Traditional way of identifying paddy plant diseases by human eyes is very time consuming and labour intensive. Hence it is necessary to set up paddy rice disease identification system to replace this traditional way. This disease classification system emphasizes on the usage of image processing technique and machine learning. This research work focuses on identifying three main types of paddy leaf diseases; Blast of Leaf, Bacterial Blight and Brown Spot. To date, there are a lot of concepts and techniques were applied to the related system. In this work, the paddy leaf disease identification system uses two different modules as for performance comparison. One is using Scikit-Learn module with K-Nearest Neighbour (KNN) algorithm; another is by using Tensor Flow with Keras model. KNN algorithm implements supervised algorithm to save all the training cases and assigns classifier to new cases based on measurement of similarity. Keras is a high level Application Programming Interface (API) running in Tensor Flow that focus in deep learning. Keras compiles, constructs and improves neural network model during training epochs. Paddy leaf disease identification system comprises of training images acquisition unit, image feature extraction unit, and neural model training unit. Upon completion of training on databases of related images, both systems are tested with test samples image to perform classification of the aforementioned three types of paddy leaf diseases. The Google Colab Notebook is used for the two algorithms realization and simulation. Python script can be ran on smart devices to load the KNN data file or Keras model files and perform on-site paddy leaf disease identification. When running on these smart devices, the input image can be chosen from either camera or photo album. The Python script is able to execute the disease identification on the smart device with correct output results. The performance of paddy leaf disease identification system by KNN algorithm and Keras model is evaluated and compared from the aspects of computational time and samples test for both models. Simulation results showed that the three major paddy leaf diseases were successfully recognized by using both models. In term of accuracy of disease detection, it can be further enhanced if more inputs of labelled images are feed as training data. For the computational time, the proposed system with the KNN algorithm able to complete all tasks within 4mins 13sec, while the Keras model required total processing time of 44mins 35sec. In terms of computational accuracy, the KNN algorithm achieved up to 81.33%, whereas the accuracy of the paddy leaf disease identification system by the Keras model is 74.67%. Thus it is clearly showed that the KNN algorithm performed better for the case of paddy leaf diseases identification system.