Soil classification using active contour model for efficient texture feature extraction

Abstract

Precision farming is a systematic approach in agriculture that aims in improving economic and environment status of the farmers. It is achieved by having prior knowledge on soil texture, nutrient, pH and other climatic conditions. Hence this paper proposes a soil classification for crop prediction approach that uses an active contour algorithm for band estimation in Fourier domain for efficient texture feature extraction. This approach initially segments the soil sample and extracts into the color and texture features. The approach proposes a texture feature extraction where the image is initially transformed to Fourier domain of a 2D-discrete Fourier transform. The image in the Fourier domain is classified into high and low-frequency bands. The cut off frequency is decided by final contour of active contour method, where initial circular contour is used for estimating final contours on Fourier coefficients. This leads to the estimation of an irregular-shaped cut off frequency along with the 2D Fourier coefficients, instead of using a circular-shaped cut off frequency. A local binary pattern (LBP) from the high-frequency band image extracts texture feature. The extracted texture and color features are trained using a fully connected network. Active contour-based proposed model was evaluated by metrics F1-score, accuracy, specificity, sensitivity, and precision on soil datasets of Kaggle and IRSID. The accuracy, F1-score, specificity, precision, and sensitivity of proposed approach active contour-based were estimated as 97.89%, 97.87%, 99.46%, 98.11 and 97.94% respectively when evaluated in the Kaggle dataset. The evaluation results of proposed active contour model based soil classification outperform other traditional approaches.