**7. Research and development in digital farming**

An overview of the published literature on the actual status of ICT usage in digital farming, particularly IoT-fog/edge/cloud computing, and Blockchain technologies reveals that most growers are interested in understanding the optimum conditions in open-field and closed-field crop production that results in reducing inputs, and at the same time maximized crop yield and quality. Our previous studies and survey show that some of the trending research topics in this context include (1) development of digital twin models that receives live data from various wireless sensors for improving efficiency of crop production systems [93], (2) adaptation of multi-robot platforms for wireless and IoT data collection [94], (3) health assessment, stress identification, and early disease detection using UAV remote sensing [95], (4) development of soil-test kits that can be mounted on mobile-robots for spontaneous determination of macronutrients in soil [96], (5) yield prediction and yield estimation using

model-based and AI algorithms [97–99], (6) evaluation of crop growth environment prior to the actual cultivation for preventing yield loss (i.e., predictive models that can be leveraged as a part of digital twin) [100], (7) development of virtual orchard models using photogrammetry [101], (8) smart irrigation with solar powered IoT controlled actuators [102], (9) reducing time losses of machinery and increasing their field efficiency by using fleet management software [103], and (10) robotic weeding and harvesting [104, 105]. The success of such systems in our point of view is intimately linked to some important factors like the accuracy and complexity of ML/ DL algorithms used to make IF decisions, as well as the availability of enough datasets to train and validate the ML/DL algorithms. From a Blockchain point of view, the horizontal and vertical scalability of IoT systems introduces more complexity in data sharing models within IF systems. The success of Bitcoin, as a result of Blockchain, is proven but the mutual collaboration between Blockchain contributors requires more maturity. Moreover, more efforts and works have to be provided to sensitize the public, the community of regulators, and the contributors about the need to invest in Blockchain development, without forgetting to address the scalability challenge (technologically speaking, it has a direct impact on the number of transactions). Furthermore, farmers in IF ecosystems need to make payments and receive subsidies from the government using cryptocurrency, transactions in this situation are susceptible to be targeted with selfish mining [106]. Blockchain is an open system, any miner can join the chain, and selfish miners can outperform honest miners and then can threaten the security of the transaction. It is a fact that Blockchain frameworks and updates for coding are publicly available, but they often lack the needed level of validation and verification against bugs, security breaches, and errors [107], so new researches and efforts are required in this direction.

Another important needed research is how to achieve interoperability between the Blockchain projects namely cross-chain, or between Blockchain and the exiting data models. The required interoperability in Blockchain enables users to take the full benefits of distributed Blockchain in terms of sharing information smoothly. As the main purpose of Blockchain is to fight against the centralization aspect, a big concern should be given to show how to build a strategy to share agricultural data (known crops diseases and solutions, best practices to increase yield) between farmers' decentralized ecosystems. The environmental impact of these technologies is always ignored or never addressed. Since sensors and electromagnetic fields generated by gateways are directly interacting with animals, soil, and vegetation, a serious study should be made to evaluate the degree of impact that the waste material of such technologies can have on the environment.

#### **7.1 Machine learning for IoT-based digital farming**

The efficiency and effectiveness of agriculture are driven by machine learning and deep learning techniques, these two mechanisms enable machines to learn and analyze data without even being programmed. ML/DL has emerged simultaneously with the Big data discipline to detect relationships, analyze patterns, and make predictions in farming activities. An example of applying a supervised machine learning algorithm with multiple distance detection sensors for autonomous navigation of a field agent robot is proposed by the SunBot project and shown in **Figure 11**. This robot is used for health assessment inside berry orchards and to collect data for supporting digital agriculture. Since traditional approaches and methods for farming management do not allow to increase productivity, farms nowadays need to be partially or

*Digital Agriculture and Intelligent Farming Business Using Information and Communication… DOI: http://dx.doi.org/10.5772/intechopen.102400*

**Figure 11.**

*Application of machine learning as a knowledge-based control approach for assisted navigation of a four-wheel steering field robot agent. Source: SunBot.de.*

fully automated using IoT systems to collect data, and ML/DL to make data inspections and drive the decision-making tasks. ML/DL technology helps farmers and scientists to select the appropriate species that respond to specific requirements in terms of diseases resistance, adaptation for specific aquatic or soil conditions, this classification task was quite tedious for farmers or scientists, but with ML/DL, a huge quantity of unorganized data is gathered and analyzed automatically to finally choose which genome is suitable for breeding. In some cases, such as plant health monitoring, it is needed to compare plants according to their colors, leaf morphology, and shapes, in that case, ML/DL can be the solution to perform the fast and accurate classification. In this context, Thaiyalnayaki et al. [108] used SVM to classify soybean diseases, and [109] performed plant leaf diseases classification based on visible symptoms.

Soil management is another farming process that has benefited from ML/DL and IoT technologies, the buried sensors collect real-time data about the underground ecosystems such as temperature and moisture, and transfer them to ML/DL algorithms to estimate the quantity of water needed for irrigation, or evaluate the quantity of nutrients required for optimal growth of crops. Superficial sensors play a major role in measuring temperature, humidity, pressure, evaporation, and evapotranspiration, these climatological and hydrological parameters among others can be used by ML/DL algorithms to estimate exactly how much water is needed to irrigate a given surface area without any wastage. To avoid wastage related to weather forecast uncertainty, Chen et al. [110] used a short-term weather forecasts method to propose an optimal irrigation strategy. Another important role of ML/DL in intelligent farming is the accurate yield prediction in quantity and quality, this prediction can be useful in crop monitoring tasks and market price forecasting. From this vision, many popular ML/DL algorithms are compared in Ref. [111] in terms of three crops yield prediction, they reported good prediction skills of the SVM ML algorithm compared to the other tested ML/DL methods. Traditional methods to control crops diseases widely spread pesticides in all the field, this treatment method leads to wastage and does not ensure the required level of efficiency, as well as harming of environment. Modern farms use computer vision techniques to accurately detect where to apply pesticides, when to apply, how much is needed, and use drones to apply pesticides with high precision. Consequently, more financial benefits are won by the farmer with no environmental side effects. Weeds density detection and treatment are examples of computer vision use case that was applied by [112] to control the area of treatment.

Like crops management monitoring, there is livestock management monitoring, the use of IoT and ML/DL in this farming activity enables farmers to predict the productivity of meets and eggs based on actual or past data. For example, a drone can make a scan of the field and count the number and the position of the cattle. A computer vision system with smart cameras can monitor the mental condition of cows to detect their preferred time of milking or the quantity of feeds they want, as well as the amount of nutrients in their milk using sensors. The visible symptoms detected through computer vision techniques are used to measure animal welfare by monitoring the health conditions of animals, and predicting if a member of the cattle is sick or wants to eat or to drink.
