Multi-source sensor data fusion method for agriculture

Abstract

The invention relates to the technical field of multi-source sensors, in particular to a multi-source sensor data fusion method for agriculture, which comprises the following steps of: firstly, partitioning large-area agricultural planting, acquiring different data by using various sensors, then dividing and classifying the acquired data, and performing variance calculation on the data of each set; and eliminating invalid data by using a GRBs judgment criterion, and carrying out secondary fusion on the data set after elimination processing through a BP neural network to obtain prediction data. Hierarchical data fusion is carried out on the agricultural collected data according to the geographic position and time, and different data processing methods are adopted for different layers, so that the information redundancy and time consumption are reduced, the precision and accuracy of the collected data are improved, and multi-sensor data based on the real-time position is provided for agricultural intellectualization.

Description

Technical Field

[0001] The present invention relates to the technical field of multi-source sensors, and particularly relates to a multi-source sensor data fusion method for agriculture. Background Art

[0003] At present, multi-source sensor data fusion mainly adopts methods such as DS evidence fusion algorithm and Kalman filtering. Through recursive estimation of the state space, multi-dimensional data fusion is realized. However, in the case of complex and numerous data caused by multiple different types of sensors, the lack of spatio-temporal information positioning leads to low processing efficiency and low fusion accuracy, resulting in untimely and inaccurate monitoring of the growth progress of crops, and further affecting the effectiveness of agricultural intelligence in work such as planting, irrigation, fertilization, and pest control. Summary of the Invention

[0004] The purpose of the present invention is to provide a multi-source sensor data fusion method for agriculture, which performs hierarchical data fusion on agricultural collected data according to geographical location and time. Different layers adopt different data processing methods to reduce information redundancy and time consumption, improve the accuracy and precision of the collected data, and provide multi-sensor data based on real-time location for agricultural intelligence.

[0005] To achieve the above purpose, the present invention provides a multi-source sensor data fusion method for agriculture, including the following steps:

[0006] Step 1: Set the planting areas where sensors are deployed as A, B, …, N, and represent them with corresponding sets. Each set contains different data attributes;

[0007] Step 2: Calculate the arithmetic mean of the data sets of the corresponding attributes of each area, and calculate the standard deviation of the samples;

[0008] Step 3: Use the Grubbs judgment criterion to clean the invalid data of the attribute X data set to complete the first-layer fusion;

[0009] Step 4: Use the Max-Min normalization method to normalize all the data after the first-layer fusion to ensure that the data is within the same scale range;

[0010] Step 5: Repeat steps 2 to 4 until all the set data of all areas are processed;

[0011] Step 6: Use the BP neural network to perform secondary fusion on the obtained normalized data to obtain the final predicted value.

[0012] Optionally, in step 1, region A is represented by set A, A = {X, Y,..., Z}, where X, Y,..., Z respectively represent the attributes of the data set obtained by the sensor, X = {X1, X2,..., X n}, Y = {Y1, Y2,..., Y n},..., Z = {Z1, Z2,..., Z n}.

[0013] Optionally, in step 2, the expression for calculating the average value of attribute X is:

[0014]

[0015] The expression for calculating the standard deviation of sample X is as follows:

[0016]

[0017] where n is the size of the data volume of the corresponding attribute data set.

[0018] Optionally, in step 3, given a significance level α = 0.05, look up the table to find the critical value g0(n, α) of the Grubbs statistic,[[]] is a small probability event. Assume that the measured value of sensor i is X i , calculate its , if , then the data X i is valid, otherwise it is invalid and is excluded;

[0019] Among them,

[0020] Optionally, the expression of the Max-Min normalization method is:

[0021]

[0022] where X i is an attribute value in the set of attribute X, min{X j} is the minimum value in the set of attribute X, and max{X j} is the maximum value.

[0023] Optionally, the process of performing secondary fusion using a BP neural network includes the following steps:

[0024] Step 6.1: Input the normalized data into the network;

[0025] Step 6.2: Initialize the maximum number of iterations, initial weights, thresholds, and allowable minimum error parameters;

[0026] Step 6.3: Calculate the input values and output values of each layer;

[0027] Step 6.4: Calculate the output layer error;

[0028] Step 6.5: Determine whether the error has converged. If so, end the fusion; if not, correct the weights and thresholds and return to step 6.3 to continue.

[0029] The present invention provides a multi-source sensor data fusion method for agriculture, which first divides large-scale agricultural planting into zones and uses multiple sensors to collect different data, then divides and classifies the collected data, and then calculates the variance of each set of data; uses the Grobes judgment criterion to eliminate invalid data, and performs secondary fusion of the eliminated data set through a BP neural network to obtain predicted data. The present invention performs layered data fusion on the agricultural collected data according to geographic location and time, and uses different data processing methods for different layers, thereby reducing information redundancy and time consumption, improving the precision and accuracy of collected data, and providing multi-sensor data based on real-time location for agricultural intelligence. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0031] Figure 1 It is a schematic diagram of fusion steps of a multi-source sensor data fusion method for agriculture according to the present invention.

[0032] Figure 2 A schematic diagram of the regional division of the present invention;

[0033] Figure 3 This is the BP neural network processing flow. DETAILED DESCRIPTION

[0034] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present invention, and should not be construed as limiting the present invention.

[0035] The present invention provides a multi-source sensor data fusion method for agriculture, comprising the following steps:

[0036] S1: Set the planting area where the sensor is deployed as A, B, ..., N, and represent it with corresponding sets, each of which contains different data attributes;

[0037] S2: Calculate the arithmetic mean of the datasets corresponding to the respective attributes of each region, and calculate the standard deviation of the samples;

[0038] S3: Use the Grubbs' criterion to clean the invalid data in the dataset of attribute X, and complete the first-level fusion;

[0039] S4: Use the Max-Min normalization method to normalize all the data after the first-level fusion to ensure that the data is within the same scale range;

[0040] S5: Repeat steps S2 to S4 until all the aggregated data of all regions is processed;

[0041] S6: Use the BP neural network to perform secondary fusion on the obtained normalized data to obtain the final predicted value.

[0042] The entire region classification and data fusion process is as Figure 1 shown. Specifically:

[0043] In step S1, region A is represented by set A, A = {X, Y,..., Z}, and the datasets obtained by different sensors are represented by attributes X, Y,..., Z, X = {X1, X2,..., X n}, Y = {Y1, Y2,..., Y n},..., Z = {Z1, Z2,..., Z n};

[0044] Figure 2 It is a schematic form of stratifying and recording each region by geographical location or time.

[0045] In step S2, calculate the arithmetic mean of the datasets corresponding to the respective attributes of each region. For example, the mean of attribute X in region A:

[0046]

[0047] And calculate the standard deviation of sample X:

[0048]

[0049] where n is the size of the data volume of each attribute dataset.

[0050] In step S3, use the Grubbs' criterion to clean the invalid data in the dataset of attribute X. The formula is as follows:

[0051]

[0052] At a given significance level (α = 0.05), look up the critical value g0(n, α) of the Grubbs statistic in the table. is a small probability event. Assume the measured value of sensor i is X i , and calculate its . If , then the data X i is valid; otherwise, it is invalid and should be excluded.

[0053] Step S4: After the first - layer fusion, use the Max - Min normalization method to normalize all the data after the first - layer fusion to ensure that the data is within the same scale range.

[0054]

[0055] In step S5, repeat steps S2 to S4 for different attributes in other regions.

[0056] In step 6, use the BP neural network (the processing flow is as Figure 3 shown) to perform secondary fusion on the normalized data obtained in step S5 to obtain the final predicted value, and dynamically analyze and monitor the real - time growth status of crops through the predicted data.

[0057] Furthermore, the present invention also proposes specific embodiments for auxiliary explanation. The specific implementation steps are as follows:

[0058] (1) Divide and classify the data collected from N planting areas in the vineyard. The set A = {X: temperature, Y: humidity,..., Z}, where the attribute temperature: X = {X1, X2,..., X n}}, the attribute humidity: Y = {Y1, Y2,..., Y n}}, the attribute Z = {Z1, Z2,..., Z n}}. Among them, the data collected for the temperature attribute in area A is X = {X1 = 30.2, X2 = 30.7, X3 = 30.5, X4 = 30.1, X5 = 20.9, X6 = 30.4, X7 = 35.7, X8 = 30.6}, and the data collected for the humidity attribute is Y = {Y1 = 0.55, Y2 = 0.50, Y3 = 0.57, Y4 = 0.70, Y5 = 0.52}.

[0059] (3) Calculate the arithmetic mean of each attribute in set A Calculate the temperature: Calculate the humidity: And calculate the standard deviation S of each attribute according to the mathematical statistics formula X , S Y ,.., S Z , The standard deviation S of the calculated temperature x = 4.0503, and the standard deviation S of the humidity Y = 0.0785;

[0060] (4) Using the Grubbs' judgment criterion, the temperature attribute is calculated , given the significance level (α = 0.05), n is 8, and the critical value of the Grubbs' statistic is found by looking up the table ; for the humidity attribute , given the significance level (α = 0.05), n is 5, and the critical value of the Grubbs' statistic is found by looking up the table . P(g i ≥ g0) = α is a small probability event. If , then the data Xi, Yi are valid; otherwise, they are invalid. Therefore, X5 and X7 in the temperature attribute X and Y4 in the humidity attribute Y are removed;

[0061] (5) After the first fusion of (1)-(4), the remaining data are uniformly normalized using the Max-Min normalization method. For the attribute temperature: X = {X1 = 30.2, X2 = 30.7, X3 = 30.5, X4 = 30.1, X6 = 30.4, X8 = 30.6}, the transformation is as follows:

[0062]

[0063] Then the new sequence X' = {X'1 = 0.1667, X'2 = 1, X'3 = 0.6667, X'4 = 0, X'6 = 0.5, X'8 = 0.8333} ∈ [0, 1]. Similarly, for the attribute humidity Y, the new sequence Y' = {Y'1 = 0.0714, Y'2 = 0, Y'3 = 0.1000, Y'5 = 0.0286} ∈ [0, 1], ensuring that the data are within the same scale range for the next fusion;

[0064] (6) The normalized new sequences X', Y' are input into the BP neural network. The BP neural network calculates the input and output values of each layer and the error of the calculated temperature and humidity at the output layer. If it is less than the given value (converges), then it ends; otherwise, the output value is re-input into the BP neural network, as Figure 3 shown. The output results are shown in the following table.

[0065] Desired state Region Temperature Humidity <![CDATA[Error (10 -5 )]]> Predicted state Water shortage Region A 30.6572 55.21％ 5.457578 Water shortage Water shortage Region B 30.5519 54.98％ 23.19745 Water shortage Unclear Region N 30.7091 54.57％ -372.12 Unclear

[0066] The above-disclosed is only a preferred embodiment of the present invention. Of course, the scope of the rights of the present invention cannot be limited thereby. Those of ordinary skill in the art can understand all or part of the processes of implementing the above embodiments, and the equivalent changes made according to the claims of the present invention still fall within the scope covered by the invention.

Claims

1. A multi-source sensor data fusion method for agriculture, characterized in that, It includes the following steps: Step 1: Set the planting areas where the sensors are deployed as A, B, …, N, represented by corresponding sets, and each set contains different data attributes; Step 2: Correspondingly calculate the arithmetic mean of the data sets of the corresponding attributes in each area, and calculate the standard deviation of the samples; Step 3: Use the Grubbs' judgment criterion to clean the invalid data in the data set of attribute X, and complete the first-level fusion; Step 4: Use the Max-Min normalization method to normalize all the data after the first-level fusion to ensure that the data is within the same scale range; Step 5: Repeat Steps 2 to 4 until all the set data of all areas are processed; Step 6: Use the BP neural network to perform secondary fusion on the obtained normalized data to obtain the final predicted value.

2. The multi-source sensor data fusion method for agriculture according to claim 1, wherein, In step 1, region A is represented by set A, A = {X, Y, ..., Z}, where X, Y, ..., Z respectively represent the data sets obtained by the sensors with attributes, X = {X1, X2, ..., X n}, Y = {Y1, Y2, ..., Y n}, ..., Z = {Z1, Z2, ..., Z n}.

3. The multi-source sensor data fusion method for agriculture according to claim 2, wherein, In Step 2, the expression for calculating the mean value of attribute X is: The expression for calculating the standard deviation of sample X is as follows: where n is the size of the data volume of the corresponding attribute data set.

4. The multi-source sensor data fusion method for agriculture according to claim 3, wherein, In step 3, given a significance level α = 0.05, look up the critical value g0(n, α) of the Grubbs statistic in the table. is a small probability event. Assume the measured value of sensor i is X i , and calculate its If then the data X i is valid; otherwise, it is invalid and should be excluded. Among them, 5. The multi-source sensor data fusion method for agriculture according to claim 4, wherein, The expression of the Max-Min normalization method is: where X i is an attribute value in the set of attribute X, min{X j} is the minimum value in the set of attribute X, and max{X j} is the maximum value.

6. The multi-source sensor data fusion method for agriculture according to claim 5, wherein, The process of performing secondary fusion using the BP neural network includes the following steps: Step 6.1: Input the normalized data into the network; Step 6.2: Initialize the maximum number of iterations, initial weights, thresholds, and allowable minimum error parameters; Step 6.3: Calculate the input values and output values of each layer; Step 6.4: Calculate the output layer error; Step 6.5: Judge whether the error converges. If so, end the fusion; if not, correct the weights and thresholds and return to Step 6.3 to continue execution.

Images