Classification of Bangladeshi soil texture using convolutional neural network

Abstract

In agriculture, soil is one of the most potential output sources. That is why, if we can foresee the soil’s nature and how it will turn in the future as well as it’s other qualities, we may achieve adequate monitoring and sustainable agriculture field usage. We can forecast many soil textures using different CNN models by doing Soil classification. As a result, our major goal is to forecast it and utilize a Convolutional Neural Network (CNN) to do so. We have applied the VGG16, ResNet50, Inception V3, Xception, and VGG19 and these are a kind of algorithm that has the capability to organize a huge number of images of separate divisions. Additionally, in our research, another algorithm is used, which is deeply related to visionary purposes. The algorithms have played a significant role in image augmentation in our research. The input is turned into a set of filters in the hidden layers to construct feature maps in the CNN model. We have used more than 2000 soil images as our data set, which helped for the betterment of our research. Images of several soil samples are used to train and evaluate these models. We have also used more than 4096 soil images of Bangladesh, creating a new scope for our research. A machine vision system consisting of a smartphone camera with an external lens, elimination chamber, USB connection, and a laptop for algorithm processing activities will be used to prepare the data. In general, the current research was carried out with five goals in mind which will be discussed in further depth in the following sections. On photos of different soil samples, these models were trained and tested. With the best accuracy percentage, the suggested models could predict soil pictures. More than 90% of accuracy from each model has been obtained, except for Xception model, where we get an accuracy of 85%. In the end, this approach will be less costly and a waste of time alternative to experimental methods for classifying the kind of soil textures on a broad scale.