Abstract:
Hyperspectral imaging is a powerful technique in remote sensing that found its application in various fields such as agriculture, health monitoring, target detection, etc. Heavy metal contamination in soil and food crops has been considered a major problem for several decades. Owing to their toxicity and persistence, heavy metals in soils are one of the most hazardous pollutants in the environment. Therefore, hyperspectral remote sensing of heavy metal contamination in soils and food crops has been widely examined for both qualitative and quantitative detection. This work will explore the application of artificial intelligence and hyperspectral imaging as a robust tool towards the classification and prediction of heavy metals.
Multiple preprocessing (Savitzky Golay, Standard Normal Variate, and Multiplicative Scatter Correlation), data reduction (Principal Component Analysis and Linear Discriminant Analysis), and estimation models (a Random Forests, Support Vector Machines, and K-Nearest Neighbors) were implemented on different datasets to evaluate their effect on classification and prediction results. Three case studies were investigated on the Sorghum Plant, Salinas A-Scene (vegetation), and Paint Condition Assessment to establish an estimation and classification model based on the correlation between the selected features and the full-spectrum features respectively. Results showed that the RF model achieved the highest accuracy in comparison with the rest of the models, with accuracies of 98 %, 90 %, and 98 % for the three case studies respectively. The high accuracy of the results and predictions showed that, in combination with the appropriate spectra-preprocessing and data reduction techniques, machine learning algorithms with hyperspectral imaging and remote sensing stand out as an advanced prediction and classification system for soil and agricultural products.