Detection of chlorine in potassium chloride and potassium sulfate using chemical imaging and artificial neural network

Chlorine in potassium chloride and potassium sulfate must be detected due to its negative effect on soil. Although the laboratory-based chlorine measurement tests are reliable, they are time-consuming, expensive, and requires chemical agents and highly skilled operators. Therefore, the novelty of the present research is developing a fast, accurate, and cheap machine-based method to measure the amount of chlorine. The purpose of this research was to apply hyperspectral imaging and machine learning techniques to detect chlorine content in potassium chloride and potassium sulfate. Different percentages of chlorine in potassium chloride and potassium sulfate products were prepared with ranges of 53.1–50.05 and 1.47–2.13 %, respectively. Hyperspectral images were captured from the sample at the range of 400–950 nm. Mean, minimum, maximum, median, variance, and standard deviation features were extracted from the image channels corresponding to the effective wavelengths. The extracted features were classified using artificial neural network method and highest accuracy of the best models for potassium chloride and potassium sulfate were 95.6 and 94.4, respectively. The combination of hyperspectral imaging and machine learning promises reliable detection of chlorine content in potassium chloride and potassium sulfate in industrial systems with high speed and low cost.