Basic Integration of Artificial Intelligence of a Plant Experimentation Chamber with LEDs and Sensors through Connection to the IoT with Node-RED and Securing Access to Data ^†

Abstract

Experimental chambers play an important role in plant science and agronomy by maintaining and providing ideal conditions so that experimental data can be collected under specific and controlled conditions. To ensure that the environmental parameters inside the chamber are within the required control limits, the best solution is to be able to record and monitor them. This recording and monitoring task are carried out in order to increase the repeatability of the experiments in the future, in addition to providing researchers with real-time information about the conditions in which the trials are found under in vivo and in vitro conditions.

1. Introduction

The Raspberry Pi is a highly versatile minicomputer [1] that can interact with other devices at the same time. Internet of things, henceforth (IoT), and cloud computing are emerging technologies. IoT is a concept in which electronic devices communicate, identify, and process with each other without human interruption. Cloud computing is a technology that consumes resources from remote servers, such as virtual machines, applications, storage, and public services hosted on the Internet. IoT becomes very powerful when it converges with cloud computing.

Several research groups have used Raspberry Pi to monitor environmental conditions in the laboratory [2,3,4], plant growth chambers, and greenhouses [1,5,6,7], including detection of accurate barometric pressure, light intensity, ambient temperature, and humidity. This is used especially in combination with automated messaging to notify researchers when variables go out of range [1,8,9], which helps to minimize variations in environmental factors that can influence the experiments, and therefore improves the reproducibility and precision of the experiments. Based on these capabilities, Ghosh et al. (2018) [10] developed a special plant growth cabinet (Growcab), which uses a Raspberry Pi to help control intensity, light quality, and the photoperiod for optimize growth rate parameters. Leitch et al. (2020) [11] used the Raspberry Pi to collect data on wind speed using an anemometer. Finally, Griffiths et al. (2017–2020) [12,13] mounted underwater environmental sensors and a GPS sensor in boats to map the acoustic environment for artistic purposes. In general, the use of the Raspberry Pi seems relatively under-utilized in autonomous environmental monitoring solutions, despite its great potential. For example, Sethi et al. 2018 [14] developed a system real-time and an autonomous monitoring of ecosystems with an open, low-cost, networked device.

Chatbots [15] can be defined as a virtual assistant that simulates human behavior. Since these systems have been endowed with Artificial Intelligence (AI), they closely approximate human behavior, and their ability and fame have increased. Machine Learning (ML) integration allows us to create and train AI models.

2. Methods

For the basic integration of AI in the experimentation chamber [7], an architecture defined by Cisco System [16] as 4-level fog computing has been used. The simplified diagram of the system is shown in Figure 1. Raspberry Pi is the main node that controls our system. The sensor network is integrated by several sensors that monitor temperature, relative humidity (RH) (DHT-22 AM2302, Aosong Electronics Co., Guangzhou, China), light intensity (Lux; TSL-2561, Texas Advanced Optoelectronic Solutions Inc., Plano, TX, USA), photosynthetic active radiation (PAR, µmol m⁻² s⁻¹,) in the 400–700 nm range (SQ-421-SS SDI-12 Digital Output Quantum Sensor, Apogee Instruments, Logan, UT, USA), and an ultraviolet light (mW/cm²; UV) in the 240–370 nm range (GUVA-S12SD, Roithner Lasertechnick GmbH, Vienna, Austria). The sensors are connected to analog and digital inputs in the Arduino Uno board, whereas the Raspberry Pi is interconnected with the Arduino Uno through the serial wire type USB. The data acquired by the sensors are continuously transmitted through the Raspberry Pi to the cloud via the Internet due to its good IoT network connectivity, based mainly on Wifi and Ethernet. In addition, Raspberry Pi not only sends data to the Cloud, but also acts as a server and as a human machine interface, as will be discussed below.

The Raspbian operating system (OS) installed on the Raspberry Pi can be downloaded from the official Raspberry Pi website [17]. The installation file with the .zip extension must be unzipped to retrieve the .img file and the image must be copied to the SD card. Since October 2021, the Raspbian image (an exact copy of the operating system and content) installed on the SD card comes pre-installed with node-RED and needs to be updated. When Raspberry Pi starts up with the command sudo systemctl enable nodered.service, node-RED starts to run automatically. The access to the node-RED programming environment from the Raspberry Pi will be done by typing the IP and the address of the port (http: //127.0.0.1:1880). By default, access by user and password authentication to the source code is disabled. This authentication method represents a serious security problem for the IoT and AI applications that have been developed since they could be violated and affect the proper functioning and even the sensitivity of information and data. Therefore, it is advisable to secure access by username and password to node-RED. For this, you have to go to $ HOME/.node-red/settings.js and access the settings.js file with a text editor and uncomment the lines from 76 to 83 (Figure 2a). The password is safely encrypted using the bcrypt algorithm [18], and the hashing is carried out by writing the command node-red admin hash-pw in the terminal. It will then ask us to insert a password and will return an encrypted password that will be the one that will be inserted in the uncommented password line of the settings.js file; with this, we will have secured the node-RED network (Figure 2b).

To use the cloud services of Google Cloud [19], an account and a project in Google Cloud must be created. This platform will recognize the user by providing the authentication token that can be used for data communication from the device to the Google Sheets API. In node-RED, the GSheet [20] node will be configured, which is a basic node to write and read from these specific sheets, previously loading the library from the Linux Operating System terminal with the command npm install node-red-contrib-google-sheets. To configure the security authentication of this node, a Google service account is used following a series of steps such as: create a service account from this page, download a JSON credential object [21] for the service account, give that account access to the sheet’s API, and share the sheet with the email address of the service account, for example, [email protected].

The Telegram Bot API [15] allows the creation of programs that use Telegram messages as an interface. In addition, it allows to create your own custom Telegram clients and connect bots to our system. These are special accounts that do not require an additional phone number to set up. These accounts serve as the interface to code running somewhere on your server. The main advantage of using the Telegram Bot API and the MTProto (Mobile Transport Prococol) is that no further knowledge about complex transmission protocols is required. Thanks to this, the intermediary server will take care of all the encryption and communication with the Telegram API and communication with this server will be performed using a simple HTTPS interface.

3. Results and Discussion

If the node-RED network security process has been carried out correctly, when we access the address https://127.0.0.1:1880 where our program code is located, we will be requested to authenticate by means of username and password (Figure 3).

Once the sensors have been connected to the Arduino Uno board and the Raspberry Pi has been interfaced with the Arduino Uno, a stream of data enters through the serial port of the Raspberry Pi that is processed by the serial node which in turn is connected to the GSheet node that sends the data to the cloud (Figure 4a) and processes with the Receive Watchdog_bot node. The user writes commands to the bot in the Telegram chat (Figure 5a,c) and the Send Watchdog_bot node. The bot writes in the Telegram chat with the user and gives the information requested by the user in addition to the system alerts (Figure 4b and Figure 5b). The bot acts as a basic unit of AI as support to the user during the tests in the experimentation chamber.

The data and alerts of the system can be queried from any place in the world. The only requirement is that the device is connected to the internet and has the Telegram application installed.

As a result of the software developed, the embedded code loaded in Arduino Uno represents 12.5% (4 Kbytes) of the memory space, leaving 87.5% (32 Kbytes) of the remaining memory free. As for the program code developed and executed in node-RED, it represents less than 1% (33 Kbytes) of the memory space, leaving 99% (4 Gbytes) of the remaining memory free for other tasks.

The system has an approximated cost of Euro 600, which contrasts with other commercial solutions for environmental data logging, such as Argus (Argus Control Systems Ltd., Surrey, BC, Canada), with prices ranging from USD 10,000 to more than USD 1,000,000. Democratizing science [22] for some researchers, the allure of do-it-yourself (DIY) is a technical research challenge of building and maintaining custom equipment. However, for others, it is a financial matter, as DIY equipment at home tends to be substantially less expensive and therefore more accessible than commercial alternatives. The system cost had been assembled and configured for approximately Euro 600.

Low-cost hardware is sometimes of little use if it cannot generate reliable and replicable data. Sometimes, however, low-cost hardware is not a good option. You need to know how accurate the tool is for a given task.

The experimental chamber prototype was successfully validated in germination experiments with pea and melon seeds [7].

4. Conclusions

The proposed system has a low cost, low energy consumption, is compact, and is highly accurate to monitor the environmental variables with the appropriate sensors remotely from any place in the world. A perfect balance between accuracy and cost is achieved through the use of free, cost-effective, and open-source resources. Researchers are in an excellent position to take advantage of these tools to revolutionize the plant science area and improve reproducibility in experimentation tasks with little impact on their budgets.