Intelligent monitoring method for water conservancy irrigation based on multi-source and multi-temporal remote sensing data

Abstract

The present application relates to the technical field of neural network, in particular to an intelligent monitoring method for water conservancy irrigation based on multi-source multi-temporal remote sensing data, comprising: a labeled data set obtained by multiple data collection means and a constructed neural network; obtaining the relative error and influence degree of multi-source data according to the relevant data in the data set, the actual value of soil water content and the estimated value output by the neural network; obtaining a loss function according to the relative error and influence degree of multi-source data, and completing neural network training in combination with the data set. The present application improves the accuracy of the weight of the neural network and the stability of the structure and weight relationship, and improves the accuracy of the estimated value of soil water content.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of neural networks, and particularly relates to an intelligent monitoring method for water conservancy irrigation based on multi-source multi-temporal remote sensing data. BACKGROUND

[0002] With the continuous development of agricultural technology and automation technology, the intelligent development of water conservancy irrigation is hot. In order to realize intelligent irrigation, the irrigation area needs to be monitored during the irrigation process. The monitoring of water conservancy irrigation is mainly for monitoring the soil moisture content of the irrigation area to reflect the irrigation effect, and then the irrigation is controlled and adjusted in time. Since water conservancy irrigation involves a wide range and a wide area, in order to accurately monitor the soil moisture content, multi-source multi-temporal remote sensing data is often used for intelligent irrigation and monitoring, that is, multi-source remote sensing data is used to monitor the soil moisture content.

[0003] For monitoring the soil moisture content, the existing technology is mostly obtained according to the vegetation state or the spectral information of vegetation and soil, but a single data has great limitations in calculating the soil moisture content, so it is necessary to comprehensively judge the soil moisture content by using multi-source remote sensing data, that is, to use multiple soil moisture content indexes to give different weights to obtain comprehensive soil moisture content. However, the weights of multiple soil moisture content indexes are difficult to directly determine, so the present application uses neural network training to determine the weight relationship between multiple indexes and soil moisture content, so as to effectively use multiple soil moisture content indexes to obtain comprehensive soil moisture content. SUMMARY

[0004] The present application provides an intelligent monitoring method for water conservancy irrigation based on multi-source multi-temporal remote sensing data to solve the existing problems.

[0005] The intelligent monitoring method for water conservancy irrigation based on multi-source multi-temporal remote sensing data of the present application adopts the following technical scheme:

[0006] One embodiment of the present application provides an intelligent monitoring method for water conservancy irrigation based on multi-source multi-temporal remote sensing data, which comprises the following steps:

[0007] The first soil moisture content, the second soil moisture content and the third soil moisture content obtained according to the vegetation spectral data, the soil spectral data and the vegetation thermal infrared data are used as samples, and the actual soil moisture content, the plant coverage, the environmental illumination and the environmental temperature are used as sample labels, a neural network data set is constituted according to the samples and the sample labels, and a soil moisture content estimation value is obtained;

[0008] The difference between the environmental illumination and the optimal illumination for plant growth is recorded as an illumination factor;

[0009] The difference between the environmental temperature and the optimal temperature for plant growth is recorded as a temperature factor;

[0010] Based on the optimal light intensity, optimal temperature for plant growth, light factors, and temperature factors, the characteristics of the influence of light and temperature were obtained.

[0011] The difference between the first soil moisture content and the estimated soil moisture content is recorded as the first moisture content difference, the difference between the second soil moisture content and the estimated soil moisture content is recorded as the second moisture content difference, the difference between the third soil moisture content and the estimated soil moisture content is recorded as the third moisture content difference, and the difference between the actual soil moisture content and the estimated moisture content is recorded as the fourth moisture content difference.

[0012] The first relative error is obtained based on the first difference in water content, vegetation coverage, light influence characteristics, and temperature influence characteristics.

[0013] The second relative error is obtained based on the difference between vegetation coverage and second water content;

[0014] The third relative error is obtained based on the temperature factor, vegetation coverage, and the difference in third water content.

[0015] The first difference, the second difference, and the third difference are obtained based on the differences between the actual soil moisture content and the first soil moisture content, the second soil moisture content, and the third soil moisture content, respectively. The first weight, the second weight, and the third weight are obtained based on the proportional relationship between the first difference, the second difference, and the third difference and the sum of the first difference, the second difference, and the third difference, respectively.

[0016] The first, second, and third relative errors are fused using the first, second, and third weights, and the loss function of the current neural network is obtained based on the difference between the fused weights and the fourth water content.

[0017] Based on the dataset and using a loss function to train a neural network, the trained neural network is used to monitor soil moisture content.

[0018] Furthermore, the specific steps for obtaining the first soil moisture content, the second soil moisture content, and the third soil moisture content based on vegetation spectral data, soil spectral data, and vegetation thermal infrared data are as follows:

[0019] The first soil moisture content was obtained based on vegetation spectral data and using the anomaly vegetation index (AVI) and vegetation status index (VCI).

[0020] The second soil moisture content was obtained based on soil spectral data and using the NIR-Red-based feature space.

[0021] The third soil moisture content was obtained based on vegetation thermal infrared data and using energy balance and apparent thermal inertia.

[0022] Further, the light influence characteristic and the temperature influence characteristic are obtained according to the optimal light for plant growth, the optimal temperature for plant growth, the light factor and the temperature factor, and the specific steps include the following:

[0023] The proportional relationship between the light factor and the optimal light for plant growth is recorded as a light abnormality amplitude, and the proportional relationship between the temperature factor and the optimal temperature for plant growth is recorded as a temperature abnormality amplitude;

[0024] The sum of the light abnormality amplitude and the temperature abnormality amplitude is recorded as an abnormality amplitude factor;

[0025] The light influence characteristic is obtained according to the product of the proportional relationship between the light abnormality amplitude and the abnormality amplitude factor and the light factor;

[0026] The temperature influence characteristic is obtained according to the product of the proportional relationship between the temperature abnormality amplitude and the abnormality amplitude factor and the temperature factor.

[0027] Further, the first relative error is obtained according to the first water content difference, the vegetation coverage, the light influence characteristic and the temperature influence characteristic, and the specific steps include the following:

[0028] The sum of the light influence characteristic and the temperature influence characteristic is recorded as a comprehensive influence characteristic;

[0029] The non-vegetation coverage is obtained according to 1 minus the vegetation coverage;

[0030] The first relative error is obtained according to the cumulative result of the first water content difference, the non-vegetation coverage and the comprehensive influence characteristic.

[0031] Further, the second relative error is obtained according to the vegetation coverage and the second water content difference, and the specific steps include the following:

[0032] The second relative error is obtained according to the product of the vegetation coverage and the second water content difference.

[0033] Further, the third relative error is obtained according to the temperature factor, the vegetation coverage and the third water content difference, and the specific steps include the following:

[0034] The third relative error is obtained according to the cumulative result of the non-vegetation coverage, the temperature factor and the third water content difference.

[0035] Further, the loss function is obtained by the following specific method:

[0036] The first relative error, the second relative error and the third relative error are normalized by using a maximum-minimum normalization method to obtain a normalized first relative error, a normalized second relative error and a normalized third relative error;

[0037] The normalized first relative error, the normalized second relative error and the normalized third relative error are fused by using the first weight, the second weight and the third weight, and the fused parameter is recorded as;

[0038] The product of the fused parameter and the fourth water content difference is accumulated and averaged to obtain a loss function.

[0039] Further, the neural network has a specific network structure as follows:

[0040] The network structure of the trained and used neural network is a BP neural network.

[0041] The technical scheme of the present application has the following beneficial effects:

[0042] (1) The soil water content is comprehensively judged by using multi-source remote sensing data, which avoids the loss of some information by single data and the influence of single data by other factors, thereby improving the accuracy of data and obtaining more accurate and reliable soil water content.

[0043] (2) The weight relationship is determined by using a neural network, which avoids the instability of the weight set by human experience and obtains more accurate and stable weight relationship as much as possible, thereby improving the accuracy of the comprehensive soil water content.

[0044] (3) In the construction of the loss function, the influence of the relative error of different data sets on the final loss function is considered, the influence of the poor quality data set on the loss function is avoided, the response of the loss function to the high quality data set is improved, and the stability of the obtained neural network structure is improved. BRIEF DESCRIPTION OF DRAWINGS

[0045] In order to more clearly illustrate the technical scheme in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or the prior art description will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0046] Figure 1 The step flow chart of the present application based on multi-source and multi-temporal remote sensing data intelligent monitoring method for water conservancy irrigation. DETAILED DESCRIPTION

[0047] In order to further illustrate the technical means and effects taken by the present application to achieve the predetermined inventive purpose, the specific implementation, structure, features and effects of the water conservancy irrigation intelligent monitoring method based on multi-source multi-temporal remote sensing data according to the present application are described in detail as follows in combination with the drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. In addition, the specific features, structures or characteristics in one or more embodiments can be combined in any suitable form.

[0048] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs.

[0049] The specific scheme of the water conservancy irrigation intelligent monitoring method based on multi-source multi-temporal remote sensing data provided by the present application is specifically described below in combination with the drawings.

[0050] Please refer to Figure 1 which shows the step flowchart of the water conservancy irrigation intelligent monitoring method based on multi-source multi-temporal remote sensing data provided by one embodiment of the present application, which comprises the following steps:

[0051] Step S001: Obtain the original spectral data reflecting the multi-source soil water content in water conservancy irrigation according to the remote sensing, spectral and infrared data acquisition means.

[0052] The specific steps include:

[0053] For the monitoring system of water conservancy irrigation, the irrigation effect is mainly monitored according to the soil water content, at which time the soil water content needs to be calculated according to the remote sensing data.

[0054] The single index in the calculation of soil water content is not highly reliable, so multiple indexes need to be comprehensively judged, which specifically involves the soil water content based on the vegetation state, soil spectrum and vegetation temperature, so when collecting data, the unmanned aerial vehicle needs to be used to carry the spectrometer to collect the remote sensing, spectral and thermal infrared corresponding vegetation spectral data, soil spectral data and vegetation thermal infrared data at multiple regions and multiple times as the original spectral data of multi-source soil water content.

[0055] In the judgment process of soil water content, the unilateral characteristics are not stable enough and are easily affected by other characteristics, so multiple characteristic parameters are often used to comprehensively judge the soil water content at the present stage, so the original spectral data corresponding to the remote sensing, spectrum and thermal infrared are used to obtain multiple soil water content estimation values. The soil water content characteristics involved in the present embodiment are:

[0056] (1) The first soil water content FA based on the vegetation state.

[0057] Under the same temperature and light conditions, the growth of vegetation is mainly restricted by the soil water content. The change of soil water content directly affects the growth of vegetation, and the existing remote sensing vegetation index can reflect the growth of vegetation. Generally, the anomaly vegetation index (AVI) and vegetation condition index (VCI) obtained by using vegetation spectrum data are used to estimate the soil water content, so the anomaly vegetation index (AVI) and vegetation condition index (VCI) based on the remote sensing vegetation index reflecting the growth of vegetation are used to estimate the first soil water content FA.

[0058] (2) The second soil water content FB based on spectral characteristics.

[0059] Generally, the red spectrum and near-infrared spectrum in the soil spectrum data and the NIR-Red feature space are used to determine the soil water content, so the difference between the red spectrum and the near-infrared spectrum in the soil spectrum data can be used to describe the vegetation, the degree of drought on the ground, and the soil water content, and the second soil water content FB is estimated.

[0060] (3) The third soil water content FC based on the thermal infrared method.

[0061] Assuming that the drought event causes the soil water content to decrease, which directly affects the growth of vegetation, causing the temperature of the vegetation canopy to rise. Generally, according to the vegetation thermal infrared data and based on the energy balance model, the apparent thermal inertia (ATI) is used to obtain the soil water content, so the apparent thermal inertia (ATI) can be used to estimate the third soil water content FC according to the vegetation thermal infrared data and based on the energy balance model.

[0062] The above different characteristic quantities of soil water content always have irresistible errors, so it is necessary to integrate multiple characteristic quantities to obtain the final soil water content. At this time, the weight of different characteristic quantities directly affects the accuracy of the final soil water content, but the weight of multiple characteristic parameters is not easy to obtain directly, so the BP neural network is used in the embodiment to train the BP neural network by using the relationship between different characteristic quantities and the actual soil water content, so as to obtain a network structure containing the weight of multiple characteristics, and then the soil water content is directly outputted after the trained BP neural network inputs multiple characteristic quantities.

[0063] Therefore, in order to train the BP neural network subsequently, firstly, the first soil moisture, the second soil moisture and the third soil moisture obtained by the vegetation spectrum data, the soil spectrum data and the vegetation thermal infrared data obtained at the same time in the same region are taken as a sample, and through the real-time monitoring of the corresponding region of each sample, the environmental illumination, the environmental temperature, the actual soil moisture and the plant coverage rate of the corresponding region are obtained, the artificial label of the sample data obtained at the same time in the same region is set as the actual environmental illumination, the environmental temperature, the soil moisture and the plant coverage rate measured artificially at the same time in the same region, and according to the data set of the BP neural network composed of a large number of sample data with artificial labels, the subsequent steps are used to comprehensively judge the soil moisture by multiple characteristic quantities.

[0064] Each sample in the data set corresponds to the first soil moisture, the second soil moisture and the third soil moisture obtained by the vegetation spectrum data, the soil spectrum data and the vegetation thermal infrared data obtained at the same time in the same region, and each sample has the artificial label about the environmental illumination, the environmental temperature, the actual soil moisture and the plant coverage rate.

[0065] In step S002, the relative error of the multi-source soil moisture is obtained according to the relationship between the multi-source soil moisture data and the soil moisture estimation value in the data set.

[0066] The specific steps include:

[0067] In the process of calculating the weight of different characteristic quantities by the BP neural network, the loss function, that is, the output weight, comprehensively judges the error between the soil moisture and the actual soil moisture, and the loss function directly affects the accuracy of the finally obtained weight. The relative error of the characteristic quantity in the data set affects the reliability of the data set, and the higher the reliability of the data set, the greater the influence on the final loss function. In this embodiment, the contribution of different data sets to the loss function is obtained by representing the reliability of the corresponding data set reflected by the characteristic quantity, and a more reasonable loss function is obtained.

[0068] The monitoring of the water irrigation system mainly aims at the difference of the soil water demand or the soil moisture before and after irrigation or in the irrigation process, and reflects the irrigation demand or the irrigation effect. Therefore, in the monitoring of the irrigation system, the soil moisture needs to be determined by using multi-source remote sensing data.

[0069] The current scene mainly determines the relationship between the characteristic quantity FA, FB and FC of the multi-source soil moisture and the actual soil moisture F. At this time, the data set of the corresponding BP neural network is a large number of first soil moisture FA, second soil moisture FB and third soil moisture FC sample data, and the specific characteristics of FA, FB and FC have greater influence on the training effect of the BP neural network model by the loss function. Therefore, the loss function is determined firstly.

[0070] The relationship between the loss function and the dataset:

[0071] The current BP neural network is mainly to obtain the relationship between the feature quantities FA, FB, FC and the actual soil water content F, so as to estimate the soil water content through the feature quantities FA, FB, FC, and the loss function is the error between the estimated value of the soil water content and the actual soil water content. However, for different datasets, the data characteristics also reflect the error relationship of different feature quantities, and the error of the feature quantity directly affects the accuracy of the final soil water content estimate, so the data characteristics of different data in the dataset affect the relationship between the feature quantity and the loss function, mainly manifested as the relative error based on different feature quantities.

[0072] First, in order to obtain more accurate soil water content estimates using BP neural networks, the above dataset contains a large number of samples, where the ith sample contains first, second, and third soil water content obtained from vegetation spectrum data, soil spectrum data, and vegetation thermal infrared data. The three soil water content data contained in the ith sample are input into the BP neural network, and the output result of the neural network is the soil water content estimate of the ith sample.

[0073] In addition, according to the differences between the first, second, third, and actual soil water content and the soil water content estimate output by the neural network, they are recorded as the first, second, third, and fourth soil water content differences.

[0074] Then, calculate the relative error based on different feature quantities:

[0075] First relative error:

[0076] For the first soil water content FA based on the vegetation state, the main basis is the relationship between the vegetation growth and the soil water content, but in reality the growth of vegetation is also affected by light and temperature, so the light and temperature in the corresponding dataset have an impact on FA, resulting in an error in the obtained soil water content estimate relative to the first soil water content FA, which is specifically expressed as:

[0077]

[0078] where the first water content difference ΔFA i represents the difference between the soil water content estimate output by the BP neural network based on the ith sample data and the first soil water content FA i of the ith sample data, the greater the value, the greater the difference between the output relative to FA i FVCi denotes the vegetation coverage corresponding to the i-th sample data in the data set, and the smaller the value, the more the soil water content FA i denotes the vegetation coverage corresponding to the i-th sample data in the data set, and the smaller the value, the more the soil water content FA i denotes the non-vegetation coverage. The light factor AH i denotes the difference between the ambient light and the optimal light H0 for vegetation growth corresponding to the i-th sample data in the data set, and the temperature factor AT i denotes the difference between the ambient temperature and the optimal temperature T0 for vegetation growth corresponding to the i-th sample data in the data set, respectively denote the abnormal amplitude of light and temperature, reflecting the influence degree on vegetation generation, and then denoted as an abnormal amplitude factor; therefore denotes the influence degree of ambient light on vegetation growth, and the light influence feature AH

[0079] that is, the influence of ambient light on vegetation, and the greater the value, the greater the influence of light, so that the current FA i has a greater possibility of error; similarly, the temperature influence feature AT denotes the influence of ambient temperature on vegetation, and the greater the value, the greater the influence of temperature, so that the current FA i has a greater possibility of error; and then

[0080] is a comprehensive influence feature. DA i that is, the error between the soil water content estimation value of the i-th sample data in the data set and the first soil water content FA i of the i-th sample data.

[0081] Second relative error:

[0082] The second soil water content FB obtained for the spectral feature is mainly based on the different characteristics of different water content soils in the spectrum, but in practice, the more lush vegetation may affect the soil spectral characteristics, so the vegetation in the data set data causes FB to have an error, resulting in an error in the estimated value of the obtained soil water content relative to the second soil water content FB, which is specifically expressed as:

[0083] DB i = FVC i x AF i

[0084] wherein, FVC i denotes the vegetation coverage corresponding to the i-th sample data in the data set, and the greater the value, the greater the influence degree of vegetation growth on the spectrum, so that FB i has a greater possibility of error; DBi Error between the soil moisture content estimation value of the i th sample data in the data set and the second soil moisture content FB of the i th sample data i Second moisture content difference ΔFB i Difference between the soil moisture content estimation value output by the BP neural network for the i th sample data in the data set and the second soil moisture content FA i The greater the value, the greater the error between the soil moisture content estimation value of the i th sample data in the data set and the second soil moisture content FB

[0085] FB i .

[0086] Third relative error:

[0087] It is known that FC represents the soil moisture content reflected by the vegetation canopy temperature. At this time, the current data corresponds to the coverage of the vegetation, which affects the degree of response of the vegetation canopy temperature to the soil temperature, and the environmental temperature directly affects the reliability of the vegetation canopy temperature. Therefore, there is an error between the vegetation coverage and the environmental temperature affecting FC, which leads to an error between the obtained soil moisture content estimation value and the third soil moisture content FC. Specifically, it is expressed as:

[0088] DC i = (1 - FVC i ) × ΔT i × ΔFC i

[0089] Where FVC i represents the vegetation coverage corresponding to the i th sample data in the data set. The smaller the value, the smaller the degree of expression of the vegetation canopy temperature to the soil moisture content, so the greater the possibility of error of FC i ; temperature factor ΔT i represents the difference between the current temperature corresponding to the i th sample data and the optimal temperature T0 for the growth of the vegetation. The greater the value, the greater the influence on the vegetation canopy temperature, so the smaller the degree of expression of the vegetation canopy temperature to the soil moisture content, so the greater the possibility of error of FC i ;

[0090] DC i Error between the soil moisture content estimation value of the i th sample data in the data set and the third soil moisture content FC of the i th sample data i Third moisture content difference ΔFC i Difference between the soil moisture content estimation value output by the BP neural network for the i th sample data in the data set and the third soil moisture content FC i The greater the value, the greater the error between the soil moisture content estimation value and FC i .

[0091] In addition, the first relative error DA obtained i Second relative error DB i and the third relative error DC i The first relative error DA′ after normalization is obtained by processing the data using the min-max normalization method. i The normalized second relative error DB′ i and the normalized third relative error DC′ i .

[0092] Step S003: Construct a loss function based on the relative error of the soil moisture content characteristics in the dataset.

[0093] The specific steps include:

[0094] Based on the above steps, the relative error of soil moisture content is obtained using data features, which reflects the relative error of the BP neural network output relative to different feature quantities. Since the relative error of different feature quantities directly reflects the quality of the corresponding feature quantity in the dataset, different feature quantities contribute differently to the loss function, thus affecting the loss function.

[0095] The current loss function is intended to reflect the error between the output of the BP neural network and the actual soil moisture content of the corresponding data. A larger error results in a larger loss function. However, in actual datasets, the relative error between the estimated soil moisture content obtained using the BP neural network and the dataset features reflects the reliability of the original dataset. Higher dataset reliability has a greater impact on the loss function. Therefore, the final loss function is expressed as follows:

[0096]

[0097] Among them, F i F′ represents the actual soil moisture content corresponding to the i-th sample data in the dataset. i This represents the estimated soil moisture content obtained from the i-th sample in the dataset using a BP neural network, and the fourth moisture content difference |F i -F′ i | represents the difference between the actual soil moisture content and the estimated soil moisture content. The larger the value, the larger the corresponding output error of the BP neural network, i.e., the larger the loss function; DA′ i DB′ i and DC′ i Let F and F represent the first soil moisture content corresponding to the i-th sample data in the dataset, respectively. i Second soil moisture content FB i and the third soil moisture content FC i relative error DAi 、

[0098] DB i and DC i respectively normalized values; a i ×DA′ i +b i ×DB′ i +c i ×DC′ i indicates the error of the i-th sample data itself, the smaller the value, the more reliable the error of the estimated soil moisture value, so |F i -F′ i | contributes more to the final loss function, that is The error of all data sets data together constitutes the loss function loss of the current BP neural network.

[0099] Where α i , b i , c i respectively represent the influence of DA′ i , DB′ i , DC′ i on the error of the i-th sample data in the data set itself, which is represented as the difference between the corresponding feature quantity and the actual soil moisture content. The greater the difference, the greater the influence of the relative error of the corresponding feature quantity on the error of the data itself, which is specifically represented as:

[0100] The first weight:

[0101]

[0102] Where, the first difference |F i -FA i | represents the difference between the i-th sample data in the data set and the actual soil moisture content FA i , the greater the value; the second difference |F i -FB i | represents the difference between the i-th sample data in the data set and the actual soil moisture content FB i , the greater the value; the third difference |F i -FC i | represents the difference between the i-th sample data in the data set and the actual soil moisture content FC i , the greater the value; the greater the influence of FA i corresponding DA′ i on the error of the data itself. |(F i -FA i )+(F i -FB i)+(F i -FC i | represents the first soil moisture FA i , the second soil moisture FB i , the third soil moisture FC i and the overall difference between the data and the actual soil moisture, at this time | represents the difference of |F i -FA i | relative to

[0103] | represents the difference of |F i -FA i )+(F i -FB i )+(F i -FC i )|, the greater the value, the greater the relative error of the feature quantity FA i

[0104] DA′ i , the greater the weight.

[0105] Similarly, b i and c i are respectively represented as:

[0106] Second weight:

[0107]

[0108] wherein the first difference |F i -FA i | represents the difference between the first soil moisture FA i corresponding to the i-th sample data in the data set and the actual soil moisture, the greater the value; the second difference |F i -FB i | represents the difference between the second soil moisture FB i corresponding to the i-th sample data in the data set and the actual soil moisture, the greater the value; the third difference |F i -FC i | represents the difference between the third soil moisture FC i corresponding to the i-th sample data in the data set and the actual soil moisture, the greater the value; FB i corresponding to DB′ i , the greater the influence of the data itself error. |(F i -FA i )+(F i -FB i )+(F i -FC i )| represents the first soil moisture FA​i , the second soil water content FB i , the third soil water content FC i the overall difference between the first soil water content FA , the second soil water content FB i , and the third soil water content FC i , relative to the actual soil water content

[0109] , the second soil water content FB i , and the third soil water content FC i , relative to the actual soil water content i , the second soil water content FB i , and the third soil water content FC i , relative to the actual soil water content i | of the feature quantity FB i , the second soil water content FB

[0110] DB i , the second soil water content FB

[0111] The third weight:

[0112]

[0113] , the second soil water content FB i , and the third soil water content FC i , relative to the actual soil water content i , the second soil water content FB i , and the third soil water content FC i , relative to the actual soil water content i , the second soil water content FB i , and the third soil water content FC i , relative to the actual soil water content i , the second soil water content FB i , and the third soil water content FC i , relative to the actual soil water content i , the second soil water content FB i , and the third soil water content FC i , relative to the actual soil water content i , the second soil water content FB i , and the third soil water content FC i , relative to the actual soil water content i , the second soil water content FB i , and the third soil water content FC i the overall difference between the first soil water content FA , the second soil water content FB i , and the third soil water content FC i| relative to

[0114] | (F i -FA i )+(F i -FB i )+(F i -FC i |The greater the value, the greater the relative error of the feature quantity FC i

[0115] DC′ i The greater the weight, the greater the degree of influence of the error of its data itself.

[0116] In addition, in the first weight, the second weight and the third weight, the constant added to the numerator and the denominator is to avoid the case that the denominator is 0, and to ensure that the sum of the three weights is 1.

[0117] At this point, the loss function loss is obtained.

[0118] Step S004, complete the training of the BP neural network, and realize the multi-index comprehensive judgment of soil water content.

[0119] After the above steps, the data set data sample with artificial labels is taken as input, the loss function loss obtained in the above is combined, and the random gradient descent algorithm is used to train the BP neural network. At the minimum of the loss function, the soil water content estimation value is output, that is, the BP neural network training is completed.

[0120] At this time, for the current collected multi-source remote sensing data, the multi-source feature quantity of the collected data is directly input into the trained BP neural network, and then the comprehensive judgment of the soil water content estimation value is output.

[0121] The intelligent monitoring of water irrigation mainly aims at the extraction of the actual irrigation area and the judgment of the irrigation effect. Therefore, through the estimation of the soil water content at multiple time points during irrigation, the irrigation effect can be understood in time, the irrigation amount and the irrigation area can be adjusted, and efficient and reliable water irrigation can be realized.

[0122] The above only describes the preferred embodiments of the present application and does not limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. An intelligent monitoring method for water conservancy irrigation based on multi-source and multi-temporal remote sensing data, characterized in that, The method comprises the following steps: The vegetation spectrum data, the soil spectrum data and the vegetation thermal infrared data are used to obtain first soil water content, second soil water content and third soil water content as samples, and the actual soil water content, the plant coverage, the environmental illumination and the environmental temperature are used as sample labels, a neural network dataset is constructed according to the samples and the sample labels, and the samples are input into the constructed neural network to obtain soil water content estimation values; The difference between the environmental illumination and the optimal illumination for plant growth is recorded as an illumination factor; The difference between the environmental temperature and the optimal temperature for plant growth is recorded as a temperature factor; The optimal illumination for plant growth, the optimal temperature for plant growth, the illumination factor and the temperature factor are used to obtain illumination influence characteristics and temperature influence characteristics; The differences between the first soil water content, the second soil water content, the third soil water content, the actual soil water content and the soil water content estimation values are recorded as first water content difference, second water content difference, third water content difference and fourth water content difference; The first relative error is obtained according to the first water content difference, the vegetation coverage, the illumination influence characteristics and the temperature influence characteristics; The second relative error is obtained according to the vegetation coverage and the second water content difference; The third relative error is obtained according to the temperature factor, the vegetation coverage and the third water content difference; The differences between the actual soil water content and the first soil water content, the second soil water content and the third soil water content are respectively obtained as first difference, second difference and third difference, and the first difference, the second difference and the third difference are normalized to obtain first weight, second weight and third weight; The first relative error, the second relative error and the third relative error are fused by using the first weight, the second weight and the third weight, and a loss function of the current neural network is obtained according to the fused value and the fourth water content difference; The neural network is trained according to the dataset and by using the loss function, and the trained neural network is used to realize soil water content monitoring; The first soil water content, the second soil water content and the third soil water content are obtained by the following method: The first soil water content is obtained according to the vegetation spectrum data and by using the NDVI and the VSI; The second soil water content is obtained according to the soil spectrum data and by using the NIR-Red based feature space; The third soil water content is obtained according to the vegetation thermal infrared data and by using the energy balance and apparent thermal inertia; The illumination influence characteristics and the temperature influence characteristics are obtained according to the optimal illumination for plant growth, the optimal temperature for plant growth, the illumination factor and the temperature factor, and the specific steps comprise the following: The illumination abnormal amplitude is recorded according to the proportional relationship between the illumination factor and the optimal illumination for plant growth, and the temperature abnormal amplitude is recorded according to the proportional relationship between the temperature factor and the optimal temperature for plant growth; The abnormal amplitude factor is recorded according to the addition between the illumination abnormal amplitude and the temperature abnormal amplitude; The illumination influence characteristics are obtained according to the product of the proportional relationship between the illumination abnormal amplitude and the abnormal amplitude factor and the illumination factor; The temperature influence characteristic is obtained according to the product of the proportional relationship between the temperature anomaly amplitude and the anomaly amplitude factor and the temperature factor; The first relative error is obtained according to the first water content difference, the vegetation coverage, the light influence characteristic and the temperature influence characteristic, and the specific steps include the following: The comprehensive influence characteristic is obtained according to the sum of the light influence characteristic and the temperature influence characteristic; The non-vegetation coverage is obtained according to 1 minus the vegetation coverage; The first relative error is obtained according to the cumulative result of the first water content difference, the non-vegetation coverage and the comprehensive influence characteristic; The second relative error is obtained according to the vegetation coverage and the second water content difference, and the specific steps include the following: The second relative error is obtained according to the product of the vegetation coverage and the second water content difference; The third relative error is obtained according to the temperature factor, the vegetation coverage and the third water content difference, and the specific steps include the following: The third relative error is obtained according to the cumulative result of the non-vegetation coverage, the temperature factor and the third water content difference; The specific method for obtaining the loss function is as follows: The first relative error, the second relative error and the third relative error are normalized by using the maximum-minimum normalization method to obtain normalized first relative error, normalized second relative error and normalized third relative error; The normalized first relative error, the normalized second relative error and the normalized third relative error are fused by using the first weight, the second weight and the third weight, and the result is recorded as a fusion parameter; The product of the fusion parameter and the fourth water content difference is accumulated and averaged to obtain the loss function; The specific network structure of the neural network is a BP neural network.

Images