AQU-FRC Net: Automated soil prediction based on faster RCNN with aquila optimization

Abstract

In agriculture, selecting an “appropriate plant for an appropriate soil” is a crucial stage for all sorts of lands. There are different types of soil found in India. It is necessary to understand the features of the soil type to predict the types of crops cultivated in a particular soil. This leads to significant inconsistencies and errors in large-scale soil mapping. However, manually analyzing the soil type in the laboratory is cost-effective and time-consuming, yet it produces an inaccurate classification result. To overcome these challenges, a novel AQU-FRC Net (Aquila – Faster Regional Convolutional Neural Neural) is proposed for the automatic prediction of soil and recommending suitable crops based on a soil-crop relationship database. The soil images were pre-processed using a Scalable Range-based Adaptive Bilateral Filter (SCRAB) for eliminating the noise artifacts from the images. The pre-processed images were classified using Faster-RCNN, which utilized MobileNet as a feature extraction network. The classification results were optimized by the Aquila optimization (AQU) algorithm that normalizes the parameters of the network to achieve better results. The proposed AQU-FRC Net achieves a high accuracy of 98.16% for predicting soil. The experimental results demonstrate that the model successfully predicts the soil when compared to other meta-heuristic-based methods.