Convergence of the Internet of Things and Big Data in Intelligent Agricultural Process Management

Objectives of the WP:

Aggregate of data and learning of the system to weed and disease determination, nutrition and water regime, in order to forecast the main agro-technical measures for common winter wheat and vegetable crops grown in greenhouse conditions.

Internet of Things is a term we all hear more and more often. The rapid development of technology over the past two decades is a prerequisite for the promotion and spread of the Internet of Things.

Production automation through the deployment of robots and machines connected to the Internet, the use of sensors to collect data in agriculture, and the remote monitoring of patients are all examples of good practices achievable through the Internet of Things. Hundreds of world-renowned companies have already successfully transformed their businesses through the introduction of IoT platforms.

Internet of Things is increasingly used in agriculture, where the use of sensors to collect data can monitor and control the condition of crops.

Good yields and healthy harvests are a natural goal of every producer. However, the cultivation of plants depends on a number of environmental factors. According to the Industry 4.0 concept, the basis for improving the quality of crop production are intelligent systems that support the automated management of technological and management processes. With the help of artificial intelligence, it is possible to plan the activities of sowing and harvesting, calculating the frequency of irrigation, determining the time and amount of fertilization, personalizing the spraying with chemicals, monitoring the weather conditions etc. An essential factor for refining the activities for growing and caring for cultivated plants is the extraction of knowledge per unit of agricultural area and the application of intelligent means for its cultivation. This is possible with the use of personalized monitoring of plant development and differentiated care. The application of appropriate machine learning algorithms makes it possible to determine the range of necessary actions depending on the individual needs of each object in the field.

Machine Learning studies different mathematical models aiming to creation of algorithm for independent increasing of self-efficiency. Important and necessary condition for learning is the analysis of data set with enormous amount of observations /bug data/ in the chosen domain. The learning of a single artificial system aims to establish dependence in these data, which is realized by applying one or a group of several algorithms.

Automated identification of plant species in the field, their level of development and accompanying pests is possible with the help of machine learning methods. These methods can focus on forecasting the needs of plants for nutrients and water, early detection of problems in their development, environmentally controlled pest control, limiting pollution and environmental damage, increasing yields without expanding the area used etc.

One of the world main food crops is common winter wheat (Triticum aestivum). It is extremely stable due to its agritechnology and yet its cultivation depends on biotic and abiotic factors. Weeds, diseases and pests have a negative effect on crop growth. Detecting and eliminating them is extremely important for obtaining high yields.
In today’s world, the potential for eliminating the harmful factors is significantly increased. The new high-tech solutions allow a qualitative transformation of work processes in agriculture. According to the Industry 4.0 concept, intelligent, interconnected systems that support automated process management are a fundamental foundation of development. The advantages of digital technology allow modernization and precision of growing crops, as well as their care. Technology integration improves the economic index, productivity, efficiency and yield quality. Agriculture today is defined as precise and intelligent. Its main feature is the extraction of knowledge per unit of agricultural area and the application of automated means for their processing.

Smart farming is an approach to creating the right management decisions based on variable field characteristics in order to maximize the economic and environmental benefits resulting in optimal yields, reduce pollution, save resources, reduce human labor, reduce labor costs. equipment. The current challenges are related to the need for personalized monitoring in crop production and differentiated services. The real approach for distributed management of arable land is the treatment of plants as separate objects with different specific requirements – species, geographical coordinates, diseases, insects and others. Automated recognition, use of computer vision and artificial intelligence allows defining the individual needs of each object in the field, as well as the range of possible actions to be taken by the farmer.
Weeds are resistant, unpretentious and adapt to environmental conditions. They spread rapidly and are competitors for water and nutrients. According to the FAO (Food and Agriculture Organization of the United Nations), in 35% of the world losses in wheat production are due to weeds. Their variety in common winter wheat is extremely large. Weeds compete with common winter wheat at the beginning of its growth, so it is especially important to remove them before the period of early spring development. In the presence of weeds during this phase, the crops are diluted, which is a prerequisite for secondary weeding. In addition, weeds are an ideal environment for the development of diseases and pests.

The main diseases that occur in cereals, in particular in ordinary winter wheat are powdery mildew, septoria, black rust, brown rust and others.

The fight against weed associations, diseases and pests is important for the technology of growing of wheat and is of great importance for the formation of yield and grain quality. Deep knowledge of them is needed to detect all weed species. A classic practice is to go through the planted field and map it. Zone mapping is the identification of the type, composition, density of each species and the description of various specific characteristics. The results of the systematic mapping allow not only to monitor, but also to take preventive measures against pests in the planted areas.

Effective pest management in wheat crops requires an integrated system for collecting and processing differentiated information, for deciding on a specific action and for its precise implementation.

A possible way to gather information about the condition of crops without human intervention is the use of drones. Depending on the available sensors and actuators, drones can perform different tasks. If there is a camera in the equipment, the drone is able to capture individual specimens of the ecosystem at regular intervals. The data from these images can be used in several ways: collecting images of plant species in arable land, identifying wheat and different types of weeds in crops, identifying characteristic diseases and pests on wheat and assessing the consequences of their impact. An important element for weed control is their proper diagnosis, ie. establishing the composition of weed communities and the quantity of individual species. In order to determine the species composition of plants, as well as possible diseases and pests of wheat, it is necessary for the intelligent software system to be able to recognize them. A controlled machine learning algorithm is suitable for this purpose. A certain amount of labeled data needs to be applied. These are the designated training data. A machine learning algorithm finds a model in the data by approximating the relationship between the independent variables (characteristics) and the output – a dependent variable (label). Labels can be generated in one set manually or extracted from another system. Articular scenario training is focused on image recognition and such a task can be solved with the help of a neural network. Capturing and storing multiple images at different stages of plant growth is the basis for creating an electronic labeled catalog. In addition to serving as a basis for neural network training, the catalog can provide specialized information to various stakeholders.

As a result of neural network training, the drone will be able to recognize any plant species in the field, various diseases and pests. When using a GPS (Global Positioning System) in a drone, its geographical coordinates can be determined at any time. The Global Navigation Assistance System (GNSS) combines data from all visible satellites at a time using triangulation. This allows the exact location with GPS. Location information is provided in real time, which means that data sequences are generated when the drone is moving. These data are essential for mapping the arable land. Mapping is the process of linking heterogeneous factors to geographical coordinates in multilayer maps. This is possible with the help of so-called geographic information systems (GIS). They allow the input, processing, storage and visualization of spatial data. Map objects can be dynamically grouped for analysis, information and decision making. The dynamic grouping will allow to assess the density of weeds, to calculate the leaf mass, to identify areas with diseased plants and pests. This information is useful for determining the types of pesticides needed and their doses. Precise treatment with chemicals exactly where they are needed and the right dose, leading to a reduction in the harmful effects on nature.

Remote monitoring by satellite sensors with high spatial and temporal resolution, which are located in Earth orbits, are considered an opportunity to obtain data and make decisions in crop production. Sustainability in these solutions depends on the availability of sufficient quantity and quality data for monitoring, analysis and planning. High-quality data provide useful information, while poor quality leads to poor analysis and, therefore, to poor decisions. There are factors that can help evaluate data quality, such as: completeness, validity, uniqueness, consistency, timeliness, accuracy, and more. Unfortunately, there is limited research in assessing the quality of databases obtained from remote sensing in this area.

Various remote monitoring platforms are currently in use, including portable, aircraft and satellites, which can be used to collect data at different spatial, temporal and spectral resolutions. The most appropriate resolutions required for remote monitoring depend on a number of factors, including management objectives, crops and their growth stages, field size and the ability of agricultural machinery to change raw materials (fertilizer, pesticides, irrigation).

One of the most effective image recognition networks is the Convolutional Neural Network (CNN). It produces remarkable classification results and is used successfully in computer vision. CNN mimics the workings of the human brain in processing visual information. CNN is a multi-layered redirection network. Consisting of multiple neurons distributed in several layers, it trains by changing weights, includes a neuron to correct displacement, and can classify multidimensional vectors. However, unlike the multilayer perceptron, it is partially coupled, which prevents it from being converted. The most distinctive feature is the ability to build a hierarchical data model. It encodes characteristic properties of the image in its architecture, each layer representing a different level of abstraction.

Another aspect of the tasks in this work package is the use of satellite data for the purposes of intelligent crop production. Remote Earth Observation (REM) can be defined as processes or methods for obtaining information about objects and events by analyzing data collected without direct physical contact with the object being studied. The essence of the method is in the interpretation of the results of the measurement of the electromagnetic radiation, which is reflected or emitted by the object and is registered at some remote point in space. The obtained physical parameters of the radiation (intensity, spectral composition, polarization and direction of propagation) functionally depend on the biogeophysical characteristics, properties, states and spatial location of the object of study.

The most universal form of recording electromagnetic radiation is the aerospace image – aerial (aerial image), when received from aircraft, and space (orbital), when received from satellites. It is formed with the help of special equipment, most often – photographic, opto-electronic and radio-electronic, united under the common name sensors. The equipment that allows simultaneous reception of images from several spectral regions is multi-channel (multi-zone). When these regions are tens and hundreds of very narrow regions of the spectrum, it is called hyperspectral, and with different polarization of the radiation – multipolarization.

The principle of “multiplicity” is embedded in the remote observation of the Earth, ie. not one frame is used, but a series of frames differing in scale, angle, time, overview and spatial resolution (PRS), spectral range and polarization of the recorded radiation. Thanks to this, mutually complementary information about the objects and events on the earth’s surface is obtained and their diverse and in-depth study is ensured.

To meet the challenges of the 21st century, fundamental changes are needed in agriculture to ensure sufficiently healthy, safe and cheap food, sustainable use of resources, and a sustainable and competitive bioeconomy. The greenhouse industry can play an important role by providing fresh vegetables with a high content of vitamins and minerals. Greenhouses allow high yields of a small area, combined with high efficiency of resource use per unit of production. The development of the plants, the yield and the quality of the obtained production depend on the microclimate in the greenhouse (light, temperature, carbon dioxide), as well as on adequate fertilization, irrigation and plant protection. In the greenhouse production of tomatoes, mathematical models have been developed that allow predicting the behavior of the system under specific conditions and help reduce costs in studies or studies of long-term effects that are difficult to trace. The opportunities for the realization of intelligent agriculture allow more precise monitoring and reporting of all parameters of the living environment for higher efficiency of resource use.

The research team includes the following scientific organizations: Agricultural Academy – Institute of Plant Genetic Resources – Sadovo /IPGR-Sadovo/ and Maritsa Vegetable Crops Research Institute; Bulgarian Academy of Science – Institute of Information and Communication Technologies and Space Research and Technology Institute.