This section explains the Proposed Effective Agriculture fusion algorithm. The following pictorial representation in Figure 1 illustrates a smart Agricultural environment, in which the parameters to be monitored are climatic condition, moisture of soil, growth of crop, nutrition level, temperature of the land area and water level. The conversion of a normal agriculture area in to smart agriculture area is done using proposed data fusion. By means of an accurate decision that can be taken to take necessary steps in the form of controlling the agriculture resources effectively.
The Proposed Agriculture setting up Architecture comprises three important parts namely power from solar, deployed sensors and IoT platform. The data captured by the sensors will be collected, fused, processed followed by analysis before taking decision appropriately.
3.1. Deployment of Agricultural sensors
In the proposed agricultural environment, several sensors have been deployed to sense and gather the agricultural data that are finally needed to integrate by the proposed data fusion algorithm which is shown in figure 2. By doing so, the conditions of the resources of the agricultural field are being monitored continuously for the follow-up things to be done from the remote place by the farmers [11]. The deployed sensors are connected by IoT for further effective association with the data analysis and decision making of the environmental resources status. The IoT and AI techniques have been used here with the integration of data using the proposed data fusion algorithm for best decision making.
Multi sensor Data Fusion Environment: Sensors with Different agriculture variants have been used to provide information that are capable in monitoring and optimizing growth level of plants, land water stage, humidity stage, soil category and temperature point of the agriculture ground as per the varying environmental factors [19] .These sensors inspect by penetrating the soil and records resistive forces before sending them to the central node for further data integration and data fusion [12].
Data integration: The data integration process does the mixture of information collected from a variety of resources like soil databases, temperature figures, humidity information, water stage statistics, and Crop Growth records throughout diverse climate circumstances by multi-sensors[18].
Data Fusion: The Data fusion has statistics resulting from multi-sensors resources, connected to the agriculture systems and sensors deployment that helps farmers in making decision about the status of temperature, humidity, water, Crop Growth and Soil. In the process of multi-sensor data collection, multiple multi-sensor network nodes collect the state information of the same monitoring object [13]. Due to the dynamic characteristics of the multi-sensor, it is necessary to fuse the same kind of data and make corresponding correction to obtain a more reasonable data [20]. There will be an assumption that there really are n multi-sensors as overall, and the interpreted data of them at a assured moment is xi, i= 1, 2 … n. As per the fuzzy logic presumption, data interpreted of a multi-sensor can be regarded as a fuzzy set [21]. Further, the comparison stuck between the two fuzzy sets can be assessed by the closeness degree. At the k moment, the calculation formula of the distance degree of multi-sensor i and j observation figure xi (k ) and xj(k ) is as follows:
aij (k) = min[ xi (k ), xj(k )]/ max[ xi (k ), xj(k )] -------------------------------------------------------- (1)
When aij (k) < the preset threshold M and is measured that xi (k ) and xj(k ) are not alike, and aij (k) = 0 can be acquired.
At moment k, the steadiness among multi-sensor i and other multi-sensor interpretation is ri (k) and the procedure for data fusion is as follows:
X(k) = wi(k) xi (k),i= 1,2....n ------------------------------------------------------------------ (2)
Where, Wi(k) stands for the weight, and Ni(k) signifies a cluster of non-negative Numbers. Due to
wi (k)/ wj(k) = Ni (k)/ Nj(k) subsequent to the normalization processing of Wi(k), the concluding fusion outcome of all the multi-sensor observation data at the k th sampling can be achieved as follows:
---------------------------------------------------------------------- (3)
Data preprocessing: The effectiveness of characterization in the agriculture can be achieved by a variety of sensors by their data sensing, computation and communication. The data preprocessing starts by preparation of data that includes data cleaning, integration, transformation, and reduction [14]. In these, cleaning routines are used to seal in absent ideals followed by smoothing noisy data, recognizing peculiarities, and then correcting data contradictions [22]. In addition, Data implementation is the process of combining information from various causes of logical data accumulation. Finally, Data preprocessing can efficiently progress the feature of data mining and decrease data mining duration in agriculture part [17]. Data preprocessing plays a vital role as the transformation of raw data into an understandable format is done here.
Data Analysis and Decision making: In an agriculture dataset, the implementation of a training set has been build up as a model; at the same time as a validation set is to authenticate the model made. From the validation set, the data points have been taken out [15]. The first pace starts by switching ANN above a learning process to a running process based on the trained and test data before running. Secondly, the similar training data immediately used in the whole system to scrutinize the error rate is done [16]. Finally, comparing the Artificial Neural Network output with the expected result from the data has been undergone.
In ANN, the weighted associates permit data to go between layers (water, soil, humidity, temperature and crop growth) through it, where the sensor allows information from preceding level and estimates a a regression coefficient of all its net inputs (Tk):
T k = ------------------------------------------------------------- (4)
Here, n is considered as input contributions, w is the weight of the connection between the sensors used. On the other hand and l, y is the input from the sensor node l, and bi is bias values of the nodes .In this instance, the sensor node productivity Pi, a relocation function ri is then applied to the weighted values of the sensors.
Pi = ri (Tk) ------------------------------------------------------------- (5)
The dataset after that can be “learnt” by training set and testing set. The process of learning is described as the procedure of modifying the data identified with the transfer functions among cells while contrasting ANN response to observable values [23]. The Back Propagation (BP) approach is often used to develop a feed-forward neural network to eliminate inaccuracy, where inaccuracy is the discrepancy between the desired outcome and the goal values in this activation.