The Digital Agricultural Revolution. Группа авторовЧитать онлайн книгу.
centroid (mostly Euclidean Distance Metric). The algorithm then calculates the new mean value of each cluster. The term “centroid update” cluster is used to build this stage. Now that the centers have been recalculated, each observation is evaluated once more to see if it is closer to a different cluster. The cluster updated means are used to reassign all of the objects. The cluster assignment and centroid update processes are done iteratively until the cluster assignments do not change anymore (until a convergence criterion is met). That is, the clusters created in the current iteration are identical to those obtained in the prior iteration. Generally, K-means clustering is used in predicting crop yields.
Figure 1.5 Cotton leaf disease using DT algorithm.
v. Association Algorithm
Association algorithms look for patterns and links in data, as well as frequently occurring “if-then” correlations known as association rules. These restrictions are comparable to data mining rules.
1.3.1.5 Improving the Model With New Data
The final stage is to apply the model to new data and, in the best-case scenario, see how accurate and effective it becomes over time. The source of the new data will be determined by the problem to be solved.
1.3.2 Artificial Neural Network
ANNs resembles the human brain based on the principle that:
Information is processed by basic units known as neurons.
Signals are transmitted from one neuron to the next via connecting links.
Each connecting link has a weight associated with it, which amplifies the signal transmitted in a conventional neural network.
To determine its output signal, each neuron’s net input passes through the activation function.
One of the popular architectures of ANN is a Multiple-layer perceptron (MLP) which consists of input, hidden, and output layers. Multiple-layer perceptrons have been successfully trained in a supervised manner utilizing a widely used method known as the Error Back Propagation Algorithm to solve a variety of complex and diverse tasks. The input layer consists of nodes that receive information from external sources and passes this information to one or more hidden layers of computation nodes and an output layer of computation nodes. During the training phase, the output is calculated for every given input and compared with the desired output. Based on the error, the network is updated. During the testing phase, the network will calculate the output for any new input data. Each conclusion has a probability assigned to it. For the most part, ANN is thought to be a good answer to difficult situations. They solve intricate relationships between crop production and interconnected characteristics that linear systems can’t solve. Artificial Neural Networks are computer programs that simulate the functioning of the human brain. Artificial Neural Network is a task-based strategy that instructs the system to work based on an internal task rather than a computationally programmed task.
1.3.2.1 ANN in Agriculture
The major advantage of neural networks is their ability to predict and anticipate via parallel thinking. Artificial Neural Network can be taught instead of being extensively programmed. Artificial Neural Network was employed by Gliever and Slaughter [30] to distinguish weeds from crops. Maier and Dandy [31] used ANNs to forecast water resources factors. Song and He [32] combined expert systems and ANNs to forecast crop nutrient levels. Comax (COtton Management eXpert), an expert system, was effectively integrated with Gossym, a computer model, and cotton crop growth was simulated. This expert system was created to work continuously in cotton crop fields throughout the year. Comax considers three field parameters: irrigation timing, nitrogen content in the field, and cotton crop development.
1.3.3 Deep Learning for Smart Agriculture
Images make up a significant portion of the data collected by remote sensing. Images can provide a complete view of agricultural landscapes in many circumstances, and they can help with a range of problems. As a result, imaging analysis is an important research field in the agricultural realm, and picture identification/classification is done using intelligent data analysis approaches [33]. One such approach is DL. A deep neural network is a network that has numerous hidden layers, each of which refines the preceding layer’s output. Feature learning, or the automatic extraction of features from raw data, is a key advantage of DL. This architecture finds its applications in the computer vision field for image classification, object identification, picture segmentation, and so on.
Because of the more complicated models utilized in DL, which allow huge parallelization, it can tackle more complicated problems exceptionally well and quickly [34]. Many researchers used DL for fruit counting, predicting future parameters, such as yield production, soil moisture content, evapotranspiration, weed detection, weather prediction, and so on.
1.3.3.1 Data Pre-processing
A commonly used pre-processing step is image resizing to 60 × 60, 256 × 256, 128 × 128, and 96 × 96 pixels. Image pre-processing is also used to identify the region of interest through segmentation, background removal, conversion to grayscale, and so on.
1.3.3.2 Data Augmentation
Many computer vision tasks have shown that deep neural networks perform exceptionally well. However, to avoid overfitting, huge data have to be provided to perfectly model the training data. The goal of data augmentation techniques is to artificially increase the quantity of training images. By providing the model with a variety of data, it helps to improve the overall learning method and performance, as well as generalization capability. For small data sets, this augmentation method is critical for training DL models. Some of the popular data augmentation techniques are flipping, rotating, cropping, scaling, translation, Gaussian noise, color casting, and so on.
1.3.3.3 Different DL Models
Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs) are two types of DL models that are driving development in different areas including the agricultural field. Many other increasingly sophisticated architectures, such as AlexNet, VGG-16, VGG-19, ResNet, Inception-V3, DenseNet, and so on, have been developed. To apply such architectures to smaller data sets, some regularization techniques, like data augmentation, dropout, batch normalization, transfer learning, and pretraining, are implemented.
1.4 AI With Big Data and Internet of Things
Policymakers and industry leaders are turning to technology factors like Internet of Things (IoT), big data, analytics, and cloud computing to help them deal with the demands of rising food demand and climate change. Farmers can use big data to get detailed information on patterns of rainfall, fertilizer requirements, and more. This allows them to make decisions on which crops to sow for maximum profit and when to harvest. Farm yields are improved when the appropriate selections are made. Sensors have been integrated into farming equipment by companies like John Deere. This kind of monitoring can be lifesaving for big farms, as it notifies users of tractor availability, service due dates, and fuel refill warnings. In essence, this maximizes the efficiency of farm equipment while also ensuring its long-term health [38].
IoT is truly a ground-breaking modern technology due to its dynamic nature. When AI is integrated with the IoT, predictive