Authors - Mahek Viradiya, Shivam Patel, Sansriti Ishwar, Veer Parmar, Simran Kachchhi, Utsavi Patel, Hardikkumar Jayswal, Axat Patel Abstract - Moisture content identification in soil is crucial for various applications in agriculture, construction, and environmental monitoring. Traditional methods for moisture detection often involve labor-intensive processes and specialized equipment which can be invasive, time-consuming, and expensive. This study explores use of spectrometry data, acquired through multispectral sensors using visible light and near-infrared (NIR) spectrum ranging from 400-1000nm, for rapid and accurate moisture identification in soil and sand samples. The sensors leverage on-chip filtering to integrate up to eight wavelength selective photodiodes into a compact 9x9mm array, facilitating the development of simpler and smaller optical devices. The neural network model compromises of input layer, one hidden layer, and an output layer, developed using Tensor-flow and Keras libraries. It was trained using the Adam optimizer and sparse categorical cross-entropy loss function for 35 epochs with a batch size of 16. Results indicate that the neural network model and appropriate classifiers can successfully classify soil moisture levels into 4 distinct categories based on given dataset, demonstrating its potential as a cost-effective and efficient alternative to traditional soil moisture measurement techniques.
Open Zoom
