Hybrid Eco-Smart Agriculture: Integrating Digital Sensors and Computer Vision for Precision Soil Nutrient Analysis and Intelligent Fertilizer Recommendation
Undergraduate CSE thesis project — Varendra University.
| File / Folder | Purpose |
|---|---|
npk_image_processing.py |
Computer vision pipeline: reads a soil test kit vial image, extracts an HSV color feature, and estimates N, P, K concentration (mg/L) via linear interpolation against a manufacturer-calibrated color chart. |
images/ |
Raw dataset — one folder per soil sample (NPK 01, NPK 02, ...), each containing N.jpg, P.jpg, K.jpg vial photos taken in a controlled environment (same device, light, background). |
Dataset/cv_of_npk_detection.csv |
N/P/K values estimated by the computer-vision pipeline, plus Temp, Moisture, Ph, and Fertility label, for all samples. |
Dataset/eye_vision_of_npk_detection.csv |
Same samples, but N/P/K read manually by eye against the color chart — used as a comparison baseline for the CV pipeline. |
fertilizer_recommendation.ipynb |
Machine learning notebook: trains and compares Random Forest, Decision Tree, SVM, and Logistic Regression to classify soil fertility (Low/Normal/High) from Temp, Moisture, Ph, N, P, K. |
best_soil_fertility_model.joblib |
Final trained model (Random Forest, 93.33% test accuracy), saved for reuse. |
- Soil sample is mixed with water, left to settle, then reacted with N/P/K test reagents.
- Each vial's color is compared to a reference chart (values in mg/L) and also photographed in a fixed lighting/background setup.
- A computer vision pipeline (
npk_image_processing.py) converts each image to HSV, computes a0.7*Hue + 0.3*Saturationfeature, and estimates the mg/L concentration via linear interpolation against the calibration points taken from the same chart. - The resulting N, P, K values (plus Temp, Moisture, Ph) are used to train a fertility classifier (
fertilizer_recommendation.ipynb).
N, P, and K values throughout this project are reported in mg/L, matching the units printed on the physical soil test kit's reference color chart used to build the dataset and calibrate the CV pipeline. This keeps the computer-vision output consistent with the ground-truth labels it was calibrated against.
pip install -r requirements.txt
pip install pandas numpy matplotlib seaborn scikit-learn joblib
- N/P/K ground truth comes from a single commercial colorimetric soil test kit, not certified laboratory chemical analysis.
- Dataset is collected under one controlled imaging setup (same device, lighting, background) and has not yet been validated across multiple regions, soil types, or camera types.