Images Based AI Algorithm for Soil Analysis

Soil is the foundation of agriculture, directly influencing crop selection. Traditional soil analysis methods, though accurate, are costly and time consuming. This work presents a rapid, low-cost and image-based soil classification approach using a novel mathematical model named as Hand-crafted Feature-Augmented Convolutional Neural Network (HCFA-CNN). By combining image-based features like particles size, shape, structure, texture, and color with a 1-D CNN, it classifies soil into four main types Sand, Clay, Sandy Loam and Loam with 85.76% accuracy. Trained on self-captured 2750 high resolution microscopic images captured by 55Megapixel microscopic camera, taking soil samples from across 97 different locations of Punjab, this model eliminates the need for expensive lab work and opens the research domain for remaining soil types and nutrients detection. We also integrated this model with web interface and developed SoilAnalyzer application for soil classification.

Project Poster