Year: 2025
Type: article
Abstract: The analysis of traditional soil texture by chemical methods is both time-consuming and subject to environmental hazards at a high cost. The researcher outlines a new Deep YOLOv8 Segmentation approach (DYV8S), which segments soil textures in Un-Inhibited Arena Circumstances (UAC) pictures. The research focuses on improving the detection of soil textures through external factor reduction and avoids the requirement for chemical analysis. The proposed DYV8S technique involves execution through four separate processes. High-resolution UAC field images undergo a USDA-trio and YOLOv8 combination for extracting soil pixels. Semantic segmentation of non-soil pixels occurs through the application of Tierce-SegNet, UNet, and DeepLabV3+ models following their deployment for pixel removal. The implementation of manifold Gabor filters extracts enhanced texture information associated with soil classification improvement. A deep YOLOv8 model operating with adjusted hyperparameters performs pre-defined categories of soil texture classification and stitching procedures. The training involved soil samples belonging to seven different texture classes. The DYV8S model delivered a 99.8% accuracy rate at the same time, it showed a 0.05 error rate throughout the soil texture identification process. Scientists documented three performance metrics, namely 0.869 Jaccard index and 0.931 Dice Similarity Coefficient, and 0.976 Kappa statistic. The algorithm analysis established that DYV8S achieved maximum accuracy in addition to exhibiting low misclassification rates. The experimental method demonstrated high time efficiency while maintaining low expense levels and safety aspects when compared to conventional chemical techniques. Maintaining controlled environmental conditions alongside controlled illumination decreased the amount of distortion during soil image analysis. High accuracy happens when the DYV8S model successfully categorizes soil texture which beats traditional classification approaches. The integrated processing strategy of segmentation with feature enhancement and deep learning systems delivers a dependable method for soil texture classification while remaining quick and economical, which improves decision-making for crops and fertilizers.
DOI: 10.14419/z5kryf33
Authors: Latha Reddy N., Gopinath M. P.
Institution: Vellore Institute of Technology University
Topic: Smart Agriculture and AI
Subfield: Plant Science
Domain: Life Sciences
Open Access status: diamond
OpenAlex ID: W4415864807
FWCI:
Cites:
Cited by:
Related to: