A new machine vision and machine learning based approach for soil texture classification

Predicting soil texture in tillage operations is a crucial step to guarantee proper seedbed preparation. Doing so would help producer to increase the yield and quality of products. The traditional soil texture techniques are laborious, cost and energy intensive, time-consuming, and largely depend on the operator’s experiences. Hence, there is a need to design a rapid, non-invasive and on-site measurement method for the prediction of soil texture classes. In this context, this research study proposes a computer vision-based system to accelerate the prediction of soil texture. To this end, an imaging box was designed to capture soil images. The acquired images were preprocessed to reduce the errors in the classification process. The efficient features (EFs) were extracted from soil images to accurately predict soil texture types using a feature reduction technique. Artificial Neural Networks (ANN) and Support Vector Machine (SVM) were applied to classify soil images. The experimental results indicated the superiority of the EF-SVM over the EF-ANN model with the accuracy rate of ۹۸.۹۹%. As a result, the EF-SVM model was effective in predicting soil texture classes

کلیدواژه ها:

Soil texture ، Classification ، Machine learning ، Machine vision ، Artificial neural network