#one hot encoding

Abstract

Agriculture is one of the most important pillars of Bangladesh’s economy. However, due to some factors such as plant diseases, pests, climate change, the yield of the farming industry decreases, and the productivity decreases as well. The detection of plant diseases is crucial to avert the losses in the productivity and in the yield. It is not obvious to monitor the plant diseases manually as the act of disease detection is very critical. It needs a huge effort, along with knowledge of plant diseases and extensive processing times. Therefore, image processing technology is used to detect the plant disease, this is done by capturing the input image that undergoes the process and is compared with the dataset. This dataset is composed of diverse diseases of potato leaves in the image format.  This study aims to build a web application to predict the diseases of potato plants that will help farmers to identify the diseases so that they can use appropriate fungicide to get more yields. The purpose of this study is to assist and provide efficient support to the potato farmers. In this study, we propose a system that will use the techniques of image process to both analyze and detect the plant diseases using machine learning Conventional Neural Networks (CNN) with Tensorflow framework 2. The results of the implementation show that the designed system could give a successful result by detecting and classifying the potato leaf diseases and healthy plant.

Introduction

Agricultural industry is the backbone of our economy that contributes about 11.63% of GDP (BBS 2021). Potato is an important and leading crop in Bangladesh. Bangladesh is the seventh potato producing country in the world and ranks second after rice in terms of production and are the third most important food crop after rice and wheat in terms of human consumption in Bangladesh (FAOSTAT, 2020). According to the DAE statistics, about 9.61 million MT of potatoes have been produced in 2020 against the annual demand of about 6.82 million MT, bringing a surplus of 3.40 million MT despite some amounts is being exported (DAM, 2020). 

But in Bangladesh late blight is the most common and highly destructive, fungal disease in potato and annual potato yield losses due to late blight have been estimated at 25-57% (GEOPOTATO project report, 2016-2019). Plants are sensitive to diseases especially the plant leaves as symptoms of the disease appear first on the leaves. Due to the bad impacts of plant diseases on the both the economy and environment, the farmers should consider monitoring the crops in such a way that they may mitigate losses. It exists a way that is used by experts to monitor the crops which is the naked eye observation. This is a traditional method that has many constraints related to time consuming as the operation of monitoring is done manually, and it requires the presence of experts. However, lately, crop monitoring is being developed to be digital and semi-automatic, meaning that only from the symptoms that are shown on the leaf, the disease could be detected in an easier, quicker, cheaper way. Therefore, this digitalized method will be beneficial for the farmers as well since it will facilitate for them the detection of the diseases because most of the farmers do not have a sufficient background and knowledge about monitoring the crops and dealing with the variety of diseases that could affect them. There are many researcher reported, leaf dieses classification and detect is successfully possible by using image processing techniques of deep learning as well as machine learning. Different methods for machine learning and deep learning include the Support Vector Machines (SVM), Random. 

Forests (RF), K-nearest Neighbor (KNN), Artificial Neural Network (ANN), and Convolutional Neural Network (CNN), along with models such as AlexNet, GoogleNet, and Caffe are used to classify and detect to leaf dieses (Knaak et al., 2021). The report presented a machine-learning model including canny edge detection technique for edge feature extraction, grid color movement for extracting color features and local binary pattern (LBP) for texture analysis. Where the features were extracted combined to create a combined feature vector which was used for training the artificial neural network (ANN). The convolutional model is also capable of differentiating the plant leaves and recognizing rice plants and their diseases (Shrivastava et al., 2022). Potato leaf diseases were detected by using random forest classifiers where image pre-processing was done in two steps like image normalization and color space conversion where segmentation was done using thresholding HSV images in RGB color space and global feature descriptor (GFD), gray level cooccurrence matrix (GLCM), color histogram were used for extracting features. Finally, classification was done using random forest (RF) classifiers (Iqbal et al., 2020). The proposed system that we are suggesting in this paper could be used by the farmers to increase the yield with no need to consult experts. The core purpose of this proposed system is not aiming only at detecting the plant diseases using the image processing technology, but it aims also at directing the user farmer to use a mobile application in which he will upload the image and receive the type of disease infection along with a suggestion of needed pesticides. The digitalization of the agriculture field has known the intervention of the latest technologies namely the image processing. As a result, our system that is designed to be automated system is implemented using image processing technique using machine learning Convolutional Neural Networks (CNN) with Tensorflow Framework 2.