Soil Testing Using Image Processing

Le résumé fourni par la source

Precision agriculture is a globally growing practice that requires accurate soil health assessment. The health of the soil is characterized by parameters such as pH and nutrients like NPK. Soil testing is very important to predict crop yield and manage fertilizers and pesticides. Traditionally, soil testing methods were time and resource-intensive, often leading to delays. This discouraged farmers from conducting regular soil tests. We propose a machine learning-based application for rapid soil analysis. We have utilized 7000 soil images paired with their corresponding laboratory-tested results. These samples were used to train our CNN (Convolutional Neural Network) model to identify soil properties from images. The machine learning model ensures accuracy and robustness as it has been trained under various lighting conditions. Preliminary evaluations indicate an average prediction variance of 0.02 pH units, 1.5 kg/ha for N, 0.8 kg/ha for P, and 1.2kg/ha for K measured as the Mean Squared Error normalized by the range of actual nutrient values. This innovation aims to contribute to sustainable agricultural practices by making soil testing in real-time possible without the need for any extra equipment or expertise.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.