Abstract
Introduction
Nowadays, precision farming is a key topic. Because of the steady increase in the world's population, which is expected to reach 9.6 billion by 2050, it is vital to increase the productivity of agricultural land while reducing waste of water, fertilizers, pesticides and pesticide. To achieve these goals, precision agriculture aims to provide farmers with a wealth of information to optimize field management, by matching farming practices more closely to crop needs. This information is obtained exploiting satellite and weather data, and wireless sensor arrays, which combined with the use of GPS, Internet of Things (IoT) and machine learning allow the farmer to operate with both a control and predictive approach [1].
For the best use of precision agriculture, it is therefore essential to collect as much data as possible on the crop status. On the one hand, various technologies have been developed or improved to collect information directly in the field, in particular to measure the pH, the nitrogen compound concentrations and the humidity amount in the soil [2,3]. On the other hand, there is not yet a well-structured system for monitoring crop gas emissions, which together with the control of soil parameters, can lead to a comprehensive evaluation of the effective health status and growth of the crop [4]. In this work, a sensor array composed of four different sensing materials, i.e. SnO2 decorated with Ag, Pd, Pt and Au nanoclusters, were developed and investigated to selective detect five different gases commonly present or emitted by crops.
Material and Method
SnO2 nanoparticles were synthesized by means of sol-gel technique, by dissolving Sn(II) ethylhexanoate in a hydroalcoholic solution of water and 2-propanol. Afterwards, it was adopted a simple impregnation method to add the metal nanoclusters. For instance, to obtain the decoration with Au nanoclusters, SnO2 nanoparticles were kept stirred in a water solution together with AuBr3, at room temperature. The nanopowder was then calcined at 650°C for 2 hours.
The four sensing materials were then screen printed onto alumina substrates, which were equipped with a Pt heater and Au interdigitated electrodes.
The gas sensors were then tested with five different gases commonly present or emitted by crops: ethylene, isoprene, CO, methanol and ammonia. They were analyzed five different concentrations of ethylene, methanol and CO and three concentrations of isoprene and ammonia, respectively. The sensing responses were collected by thermo-activating the sensing materials at four different temperatures: 200, 250, 300 and 350°C.
The response values from the sensors, combined in a 3-dimensional point for each gas concentration, were processed through a Support Vector Machine (SVM) with a linear kernel to improve sensors selectivity. A first set was used to train the system, and a second one was used to optimize and test its performance.
Results and Conclusions
The Scanning Electron Microscope (SEM) analysis highlighted that the average size of the synthesized SnO2 nanoparticles was about 30-40 nm (Figure 1), while the X-ray Diffraction (XRD) highlighted a single crystalline phase (cassiterite). The SEM images showed that the metal nanoclusters were growth over the SnO2 nanoparticles surface, without modifying the SnO2 morphology, average size and crystalline structure (XRD) of SnO2. The weight concentration of metals in decorated samples was lower than 1% (Figure 2).
The gas sensing characterization was performed in a sealed aluminum gas chamber, by exploiting certified cylinders of target gases and mass flow controllers. The Figure 3 shows the Principle Component Analysis (PCA) for the SnO2/Pt gas sensors vs. tested gases. As it can be observed, the PCA clearly discriminates the different concentrations of gases. The Figure 4 a) and b) show the results obtained in the test set of gas sensing responses with SnO2/Pt gas sensor, after the training of the SVM with linear kernels fit, carried out to build the classification model. The position of each point of the test set was compared with the trained model and thus classified. Figure 4a) highlights that there was a perfect SVM classification (100%) of the different concentration of the tested gases (21 points) for the SnO2/Pt, with a relative low error in the estimated concentrations compared to the real ones (Figure 4b). The same data analysis was then performed for the other sensing materials: Ag, Au and Pd/SnO2. Finally, the SVM classification was also carried out by combining all the gas sensing responses of the four different gas sensors, by achieving a perfect classification of the gases analyzed together with a very low error in the estimated concentrations, compared with the concentrations injected in the gas chamber.
References
[1] Maia, R.F., Netto, I., Tran, A.L.H. Precision agriculture using remote monitoring systems in Brazil (2017) GHTC 2017 - IEEE Global Humanitarian Technology Conference, Proceedings, 2017-January, pp. 1-6. doi: 10.1109/GHTC.2017.8239290.
[2] Ruiz-Garcia, L., Lunadei, L., Barreiro, P., Robla, J.I. A review of wireless sensor technologies and applications in agriculture and food industry: State of the art and current trends (2009) Sensors (Switzerland), 9 (6), pp. 4728-4750. doi: 10.3390/s90604728.
[3] El-Shikha, D.M., Waller, P., Hunsaker, D., Clarke, T., Barnes, E. Ground-based remote sensing for assessing water and nitrogen status of broccoli (2007) Agricultural Water Management, 92 (3), pp. 183-193. doi: 10.1016/j.agwat.2007.05.020.
[4] Fabbri, B., Valt, M., Parretta, C., Gherardi, S., Gaiardo, A., Malagù, C., Mantovani, F., Strati, V., Guidi, V. Correlation of gaseous emissions to water stress in tomato and maize crops: From field to laboratory and back (2020) Sensors and Actuators, B: Chemical, 303. doi: 10.1016/j.snb.2019.127227.
Figure 1