Akademik Veri Yönetim Sistemi
ARABIAN JOURNAL FOR SCIENCE AND ENGINEERING, cilt.47, sa.8, ss.10751-10767, 2022 (SCI-Expanded)
Finding the particle size distribution is one of the main objectives in soil science as well as any other area to be worked on soil. There are many methods to accomplish this task, both traditional and technology-oriented. While traditional methods have many disadvantages, such as being dependent on the laboratory medium and expert knowledge, being carried out manually etc., technology-oriented devices on the other hand, are quite expensive due to hardware and production costs. In order to propose an optimized solution in between, an automated soil texture analyzer, supported by microcomputer and machine learning method, is proposed, which can eliminate the incomplete and error-prone disadvantages of the traditional hydrometer method, focusing only to find ratios of sand, silt and clay materials in a soil sample. In the light of a previous study, the structure and working principle of an automated soil texture estimation device, which works depending on the change of ultrasound intensity passed through the soil-water mixture in the 3D-printed measurement cup, has been explained in detail. The signals obtained from 80 soil samples, the contents of which were analyzed by the traditional hydrometer method, have been collected on the computer using the proposed device, pre-processed, and then feature extraction steps have been applied to be given as input to the machine learning methods. Rates of sand, silt and clay fractions of the soils have been predicted using Support Vector Regression (SVR), Multi-Layer Perceptron Neural Network (MLPNN) and Long Short-Term Memory (LSTM) machine learning methods. Best results are achieved using MLPNN structure in terms of Mean Absolute Error (MAE). Using MLPNN 54 soils for sand, 48 for silt, and 48 soils for clay, are found having estimation error of less than 10%. Worst results are obtained using LSTM method. There are 38 soils for sand, 38 for silt, and 41 soils for clay with MAE value below 10% with LSTM method. Statistical validations are presented with RMSE and Correlation Coefficient values and the achieved results are thought to be moderate especially considering the many of the competing studies are carried out using thousands of dollars worth laser diffraction based machines. The temperature effect has also been investigated by performing tests at varying temperatures from 9 to 38 degrees C.