Article Preview
TopI. Introduction
With the passage of time there has been an exponential increase in the population of the world. The necessity of nutritious and high-quality food is thus increasing. This led to keep a watch on overall growth in food production. There are several factors like metrological conditions, soil conditions, improper use of pesticides and fertilizers, diseases in plants are contributing for less productivity and poor quality of food (Nikhitha et al., 2019). The degrading quality of food has been adversely affected the health of humans and other living beings. Disease in fruits has been a primary reason of their less production (Park et al., 2018; Nikhitha et al., 2019). Thus, detection of diseases in the fruits at an early stage is necessary and thus has been a topic of concern for researchers. Figure 1 shows that over the time there is an exponential increase in the research. The researchers have been trying to explore the disease detection in plants, fruits and vegetables to help the farmers in improving the quality of the yield.
The technological advances have made it feasible to develop techniques which are capable of detecting the diseases at early stages. The improvements in hardware devices include the presence of high-resolution cameras and sensors (Kamilaris et al., 2018). Apart from these, over the past few years, various satellites have been able to transmit useful data for the research, by the efforts of various space agencies worldwide (Desai et al., 2016). This data can be processed by the computer experts who can apply the image processing techniques to decompose and process this image data set to obtain useful insights.
These techniques to obtain useful insights from the image data, a clean dataset is required and hence noise removal techniques were applied. The noise free data obtained is then segmented, for which different techniques were used by researchers as discussed in section 3 out of which clustering was recommended. However, its drawbacks and how soft computing can enhance it has been discussed. Other techniques apart from clustering have been used whereas in some models the need of pre-processing is eliminated by using appropriate deep learning techniques which are capable of determining features by their own.
The researchers at earlier stages used classification techniques including machine learning models out of which, Support Vector Machine (SVM) is most prominent (Kamilaris et al., 2018). The deep learning models came into existence in 2000s and were being used to solve complex problems. Image Processing is a computationally heavy problem and hence requires a highly complex model to evaluate and predict the outcomes. A precise study of these models will allow the developers to come up with an application which may surplus the yield, thus helping the farmers to plan better. The deployment of these models on image data of various fruit plants have been serving the farmers to enhance their production. We in this article intend to present a comprehensive analysis of various proposed models in order to provide insights of the best model on the basis of techniques used for image acquisition, image preprocessing, image segmentation and training model.
The objective of this article is mainly to recognize the most suitable method for deployment in real time which will enhance the quality of fruits and thus helping the society in positive way. The article is divided into different parts as follows. The second section is methodology; followed by a basic model to detect diseases in fruits with detailed analysis of each step is discussed in subsequent subsections. Critical analysis of all the articles related to the theme has also been provided in Table 1, which provides a comprehensive study of approximately 40 articles on the basis of various parameters like preprocessing technique, network model, dataset used, evaluation of the model and the fruit on which the research has been done.