Precision agriculture is an approach to farm to ensure that the crops and soil receive exactly what they need for optimum health and productivity. Precision agriculture offers the potential to automate and simplify the collection and analysis of information. It allows management decisions to be made and quickly implemented on small areas within larger fields. One of the advanced and inexpensive methods that provides a lot of information about the soil with the shortest time and lowest cost is measuring acoustic signals with cone penetrometer. The texture of the soil determines the percentage of the constituents of the mineral part of the soil such as sand, silt and clay.
An acoustic penetrometer was developed to provide an accurate method for determining the soil texture. This system uses a microphone to record the sound produced through the cone-soil communication and correlates this data with the soil texture.
Materials and Methods
An acoustic cone penetrometer (ACPT) was designed to determine if there is a relationship between the sound produced at the cone-soil interface and soil particle size. Three types of cones with angles of 30, 45 and 60 degrees and diameter of 20.27 mm and rod length of 300 mm according to ASAE standard S313.3 FEB1999ED (R2013) were used to determine the relationship between sound and soil texture and to choose the best angle. A microphone (20-20,000 Hz) was used to record the audio signals produced from the soil, suitable for fast dynamic responses. Audio signals were stored online through the oscilloscope section of Matlab software. To create the controlled vertical movement of the cones, a mechanical mechanism with electronic controllers was designed. This mechanism can be connected on the rails of the soilbin located in Urmia University. In this system, a 5 hp electric motor with a gearbox, an inverter to control the rotational speed of the electric motor and a digital ruler to record vertical movement were used. Soil samples were tested in 5 gallon bins.
Acoustic signals received from microphone were processed in the time-frequency domain using wavelet transform. In this research, Daubechi function type 3 is used to analyze acoustic signals. It is not possible to use processed acoustic signals directly in statistical analysis. For this reason, appropriate features should be extracted from them. From the 30 features of time domain signals, the most effective and main features include; SUM, Max, RMS, average, Var, kurtosis and Moment4. They were ranked using the feature selection section of WEKA 3.9.2 software in order to avoid increasing the volume of calculations and increasing processing speed and reducing errors. In order to analyze the signals related to several different tissues and finally distinguish the difference between these three types of tissues, the characteristic vector of the sub-signals should be analyzed.
Results and Discussion
To select the best type of cone using WEKA software, the number of features in d1 sub-signals was higher for the 45 degree cone, and it can be concluded that this cone has more ability to recognize the data.
The average values of characteristics in clay, loam, and sand had an increasing trend respectievely with a significant probability of 1% and 5%.
Acoustic signals for clay soil, which has a heavy texture and small particles, have a minimum amplitude, and for loamy and sandy soil, they were observed as medium and maximum, respectively. This will cause the values of the selected features of clay soil to be low, and as a result, the average values, variance, and standard deviation. They would be higher for loamy and sandy soil which have larger particles. In this way, as the size of the soil particles increases, hitting the cone wall would become heavier and would affect the frequency and amplitude of the signal, which will increase the values of the signal amplitude and, as a result, increase sum, max and mean.
Among the sub-signals, the maximum effect of soil texture type changes was related to d1 sub-signals for the 45̊ cone, and these signals had more potential to identify the soil texture type. Among the features, sum, average, VAR and RMS were significant at 1% probability levels. Therefore, these features in the mentioned sub-signal have more potential to detect the type of soil texture. Also, the effect of soil texture change on Moment and Kurtosis characteristics was significant at 5% probability levels.