SOIL QUALITY DETECTION FOR UPLAND RICE FARMING USING PRE-TRAINED CONVOLUTIONAL NEURAL NETWORK

Abstract

One of the most important factors that affect the yields of farm produce is the quality and usability of the soil. The analysis of the soil quality and the usability of rice, a staple crop in the world, have been of interest to many researchers. There is increasing research in upland rice farming, which is making it to gain more attention. The major challenge upland environment poses is soil suitability; due to the nutrient demand of rice plant. Convolutional Neural Network is a deep learning technique that is commonly applied in the analysis of visual imagery. This study therefore has used pre-trained convolutional neural networks like AlexNet, Inception V3, ResNet18 and GoogLeNet to evaluate the usability of upland soils for rice farming. Soil samples and images were taken from Delta, Niger, Ogun and Osun states to the laboratory for analysis of soil PH and texture. Based on the soil quality score associated with each soil sample, the soil images were trained and classified as either “good for rice farming” or “not good for rice farming” using the selected pre-trained neural network. The results show high classification accuracy; however, ResNet18 performed best with an accuracy of 92.8%, slightly better than GoogLeNet. This studies further modified GoogLeNet model due to its portability to come up with a convolutional neural network model with an accuracy of 97.24%. Further research that explores additional soil parameters such as topography and moisture content is recommended