Wetland biological environment collaborative online monitoring system and method

Abstract

The invention relates to the technical field of wetland monitoring, and particularly discloses a wetland biological environment collaborative online monitoring system and a wetland biological environment collaborative online monitoring method, which respectively use a color histogram and a direction gradient histogram to extract color features and HOG features of wetland image data, and introduce a deep learning algorithm to obtain a wetland biological environment collaborative online monitoring result. The color features and the HOG features of the wetland image are subjected to principal component analysis and compensation type interactive fusion between cross-domain features, so that accurate recognition of vegetation features in the wetland image and intelligent estimation of the vegetation coverage rate are achieved, and then correlation response analysis between wetland organisms and environmental factors is conducted on the basis. Through the mode, the identification precision of wetland vegetation features can be improved, and the adaptability of the model to different illumination conditions and complex backgrounds is enhanced, so that more reliable data support can be provided for wetland protection and management.

Description

Technical Field

[0001] The present application relates to the field of wetland monitoring technology, and more specifically, to a wetland biological environment collaborative online monitoring system and method. Background Art

[0002] As an important ecosystem on Earth, wetlands play an irreplaceable role in maintaining biodiversity, regulating climate, purifying water quality, and providing ecological services. However, with the continuous increase in human activities, wetland ecosystems are facing unprecedented pressure, and the monitoring of the dynamic changes in their biological environment is particularly important. At present, the monitoring of wetland biological environment mainly relies on traditional field surveys and sampling methods, which are not only time-consuming and labor-intensive, but also difficult to achieve real-time monitoring and comprehensive coverage.

[0003] In this regard, the invention patent with publication number CN118670457A proposes a method for collaborative online monitoring of wetland biological environment. It uses an optical camera to continuously capture high-resolution wetland images, and deploys multi-parameter water quality meters in the same area to obtain environmental element data related to vegetation growth. The biological parameters of the wetland images are interpreted to obtain parameters such as vegetation coverage and plant height. Then, by comparing the changes in parameters such as vegetation coverage, density, and plant height under different environmental conditions, the response of wetland vegetation growth distribution to changes in environmental factors is studied, thereby realizing collaborative online monitoring of the wetland biological environment.

[0004] However, in the prior art, when interpreting biological parameters, the wetland image pixels are classified using a BP neural network to distinguish different landform types such as wetland vegetation, mudflats and waters, and the direct linear transformation (DLT) method is used to convert the image into a bird's-eye view, and the coverage is obtained by counting the ratio of the number of pixels occupied by the vegetation canopy to the total number of pixels. Although the image pixel classification method based on the BP neural network can achieve a preliminary classification of wetland images to a certain extent, it may not be accurate enough for the complex and changeable wetland environment, especially when facing different lighting conditions, seasonal changes and other factors, which may lead to increased classification errors. In addition, when using the direct linear transformation (DLT) method to convert images into bird's-eye views, the conversion accuracy may be insufficient due to factors such as perspective distortion of the image and ground undulations, thereby affecting the accuracy of subsequent biological parameter interpretation.

[0005] Therefore, an optimized wetland biological environment collaborative online monitoring system and method is desired. Summary of the invention

[0006] In order to solve the above-mentioned technical problems, the present application is proposed. The embodiment of the present application provides a wetland biological environment collaborative online monitoring system and method, which respectively uses color histogram and directional gradient histogram to extract color features and HOG features of wetland image data, and introduces a deep learning algorithm to perform principal component analysis and compensatory interactive fusion between cross-domain features on the color features and HOG features of the wetland image, so as to achieve accurate recognition of vegetation features in wetland images and intelligent estimation of vegetation coverage, and then conduct correlation response analysis between wetland organisms and environmental factors on this basis. In this way, the recognition accuracy of wetland vegetation features can be improved, and the adaptability of the model to different lighting conditions and complex backgrounds can be enhanced, thereby providing more reliable data support for wetland protection and management.

[0007] Accordingly, according to one aspect of the present application, a method for collaborative online monitoring of a wetland biological environment is provided, which includes: synchronously acquiring wetland image data and environmental element data of a wetland; calculating vegetation coverage based on the wetland image data; and obtaining the response of wetland vegetation to changes in environmental factors based on the environmental element data and the vegetation coverage, wherein calculating vegetation coverage based on the wetland image data includes: extracting color features of the wetland image data using a color histogram to obtain a wetland image color feature coding vector; extracting HOG features of the wetland image data using a directional gradient histogram to obtain a wetland image HOG feature coding vector; performing feature principal component compensation interaction on the wetland image color feature coding vector and the wetland image HOG feature coding vector to obtain a wetland image color-HOG compensated interactive coding feature vector; and estimating vegetation coverage based on the wetland image color-HOG compensated interactive coding feature vector to obtain the vegetation coverage.

[0008] According to another aspect of the present application, a wetland biological environment collaborative online monitoring system is provided, comprising:

[0009] A wetland image color feature extraction module is used to extract the color features of the wetland image data using a color histogram to obtain a wetland image color feature encoding vector;

[0010] A wetland image HOG feature extraction module is used to extract the HOG features of the wetland image data using a directional gradient histogram to obtain a wetland image HOG feature encoding vector;

[0011] A principal component compensation interaction module is used to perform principal component compensation interaction on the wetland image color feature coding vector and the wetland image HOG feature coding vector to obtain a wetland image color-HOG compensation interaction coding feature vector;

[0012] The vegetation coverage estimation module is used to estimate the vegetation coverage based on the wetland image color-HOG compensation interactive coding feature vector to obtain the vegetation coverage.

[0013] Compared with the prior art, the wetland biological environment collaborative online monitoring system and method provided by the present application respectively uses color histogram and directional gradient histogram to extract the color features and HOG features of wetland image data, and introduces a deep learning algorithm to perform principal component analysis and compensatory interactive fusion between cross-domain features on the color features and HOG features of wetland images, so as to achieve accurate recognition of vegetation features in wetland images and intelligent estimation of vegetation coverage, and then conduct correlation response analysis between wetland organisms and environmental factors on this basis. In this way, the recognition accuracy of wetland vegetation features can be improved, and the adaptability of the model to different lighting conditions and complex backgrounds can be enhanced, thereby providing more reliable data support for wetland protection and management. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0015] Figure 1 Flow chart of a method for collaborative online monitoring of a wetland biological environment according to an embodiment of the present application.

[0016] Figure 2 Schematic diagram of data flow of a method for collaborative online monitoring of a wetland biological environment according to an embodiment of the present application.

[0017] Figure 3 This is a flowchart of step S3 in the wetland biological environment collaborative online monitoring method according to an embodiment of the present application.

[0018] Figure 4 It is a block diagram of a wetland biological environment collaborative online monitoring system according to an embodiment of the present application. DETAILED DESCRIPTION

[0019] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0020] As mentioned in the background technology above, patent CN118670457A proposes a method for collaborative online monitoring of wetland biological environment, which includes: synchronously acquiring wetland image data and environmental element data of the wetland; calculating vegetation coverage based on the wetland image data; and obtaining the response of wetland vegetation to changes in environmental factors based on the environmental element data and the vegetation coverage.

[0021] Specifically, in the existing technology, optical cameras are used to continuously capture high-resolution wetland images, and multi-parameter water quality meters are deployed in the same area to obtain environmental factor data related to vegetation growth. The vegetation coverage, plant height and other parameters are obtained by interpreting the biological parameters of the wetland images. Then, by comparing the changes in parameters such as vegetation coverage, density, and plant height under different environmental conditions, the response of wetland vegetation growth distribution to changes in environmental factors is studied, thereby realizing coordinated online monitoring of the wetland biological environment.

[0022] However, when interpreting biological parameters, the existing technology uses BP neural network to classify wetland image pixels, distinguish different types of land features such as wetland vegetation, mudflats and waters, and uses the direct linear transformation (DLT) method to convert the image into a bird's-eye view, and obtains the coverage by counting the ratio of the number of pixels occupied by the vegetation canopy to the total number of pixels. Although the image pixel classification method based on BP neural network can achieve preliminary classification of wetland images to a certain extent, it may not be accurate enough for complex and changeable wetland environments, especially when facing different lighting conditions, seasonal changes and other factors, which may lead to increased classification errors. In addition, when using the direct linear transformation (DLT) method to convert images into bird's-eye views, the conversion accuracy may be insufficient due to factors such as perspective distortion of the image and ground undulations, thereby affecting the accuracy of subsequent biological parameter interpretation.

[0023] In response to the above problems, this application proposes an optimized wetland biological environment collaborative online monitoring method based on the above scheme, which uses color histogram and directional gradient histogram to extract color features and HOG features of wetland image data respectively, and introduces deep learning algorithms to perform principal component analysis and compensatory interactive fusion between cross-domain features on the color features and HOG features of wetland images, so as to achieve accurate recognition of vegetation features in wetland images and intelligent estimation of vegetation coverage, and then conduct correlation response analysis between wetland organisms and environmental factors on this basis. In this way, the recognition accuracy of wetland vegetation features can be improved, and the adaptability of the model to different lighting conditions and complex backgrounds can be enhanced, so as to provide more reliable data support for wetland protection and management.

[0024] Specifically, the acquisition of wetland images is an important basic link in realizing the coordinated online monitoring of wetland biological environment. In the selection of sensors, devices with high-resolution imaging capabilities should be given priority. Wetland ecosystems are extremely complex and diverse, with a wide variety of plants and animals, each with unique morphological characteristics. High-resolution imaging equipment can accurately capture these subtle details, such as the delicate texture on the feathers of rare birds. The characteristics of its texture are of great significance for the species identification, individual identification of birds, and the study of their physiological functions such as flight and warmth. The leaf morphology of small aquatic plants, different shapes, sizes and vein distribution, can reflect the species, growth status and adaptation strategy of plants to the aquatic environment. The cave structure of benthic organisms, from the shape and size of the cave entrance to the complex structure inside, is closely related to the living habits, reproduction methods and roles of benthic organisms in the ecosystem. These detailed information plays a decisive role in species identification, ecological behavior research and functional evaluation of ecosystems.

[0025] In recent years, drone technology has emerged in the field of wetland image acquisition with its excellent flexibility and maneuverability, becoming an indispensable and effective tool. Using drones equipped with high-resolution cameras, regular or real-time image acquisition of wetlands can be achieved. When setting the flight altitude of drones, a careful balance needs to be made between image resolution and coverage. A lower flight altitude, such as about 50 meters, can obtain extremely high-resolution images, even up to 1 cm resolution, which allows us to clearly observe every detail of individual organisms, including the limb structure of small insects, the morphology of tiny aquatic organisms, and small ecological features, such as changes in water quality in small ponds, and the distribution of micro-wetland vegetation communities. However, the disadvantage of a lower flight altitude is that its coverage is limited, and the image area that can be collected in one flight is small, which is less efficient for monitoring large areas of wetlands. A higher flight altitude, such as 500 meters, can significantly expand the coverage area, and a large area of ​​wetlands can be covered in one flight, which is conducive to quickly grasping the overall layout of the wetland. But with this comes a reduction in resolution, about 50 centimeters, which makes it difficult to observe subtle biological features and small ecological phenomena. Therefore, for key research areas, such as habitats of rare species, which are crucial for biodiversity conservation and require precise understanding of the survival status and ecological changes of organisms therein, or ecologically fragile areas, where subtle changes in their ecological environment need to be closely monitored, the flight altitude of the drone can be set between 50-100 meters to obtain images with a resolution of 1-5 cm, thereby ensuring that individual organisms and small ecological features can be clearly observed. For large-scale wetland surveys, the aim is to quickly understand the overall pattern of wetlands, including the boundary range of wetlands, the distribution areas of different wetland types, and to identify the main ecological types and landform distribution, such as large areas of reed marshes, mudflats, and waters. The flight altitude can be increased to 200-500 meters, at which time the image resolution is about 10-50 cm. This setting can not only meet the needs of large-scale monitoring, but also obtain information of certain value.

[0026] Flight path planning is also a key link that cannot be ignored in wetland image acquisition. The use of grid or strip flight modes can ensure comprehensive and complete coverage of wetlands. When planning the flight path, the shape, boundaries, and topographic features of the wetland are all factors that need to be considered. For irregularly shaped wetlands, such as long and narrow or with many branches, flexible adjustments are required according to their unique contours. Through accurate mapping and analysis of wetland boundaries, a flight path that fits its shape is developed to avoid monitoring blind spots. At the same time, reserving a certain degree of image overlap is an important measure to ensure image quality and subsequent processing effects. Usually, the heading overlap is set at 60%-80%, and the lateral overlap is set at 30%-60%. Sufficient heading overlap can ensure that there is enough overlap between adjacent images along the flight direction, which is crucial for subsequent image stitching, making the stitched images smoother and more natural, avoiding cracks or dislocations. The lateral overlap helps to establish connections between different flight routes, facilitates three-dimensional reconstruction, and obtains height information of ground objects through stereoscopic observation, such as the height of trees in the wetland and the undulations of the terrain, thereby providing data support for a more comprehensive understanding of the wetland ecological environment.

[0027] Time selection has a profound impact on the quality and information extraction of wetland images. Wetland ecosystems have significant diurnal and seasonal changes. During the day, changes in light intensity and angle will have a significant impact on the quality of the image. The best shooting time should avoid periods of excessive or weak light. In the early morning, the sun has just risen and the altitude angle is low, and the objects will produce obvious shadows. These shadows can enhance the three-dimensional and layered sense of the image, making the terrain undulations more obvious, and help to highlight the contours of objects, such as small hills and dams in the wetland. In the evening, the sun's altitude angle is also low and the light is soft. The images taken at this time can better show the details and texture of the wetland, which is of great significance for identifying different objects and biological distribution. For wetland vegetation that relies on photosynthesis, it is extremely important to choose a season when vegetation growth is vigorous for shooting. In spring and summer, wetland vegetation grows luxuriantly, and different types of vegetation show their own unique colors, forms and distribution characteristics. Shooting at this time can better show the types, distribution and growth status of vegetation, which is helpful for classifying and studying wetland vegetation communities, understanding the succession laws of vegetation and its relationship with the wetland environment. In addition, weather conditions are also an important factor affecting image quality. Try to choose clear and cloudless weather for shooting. The obstruction of clouds will form large shadows on the image, seriously affecting the integrity and accuracy of the image, and it is impossible to clearly observe the wetland conditions under the clouds. At the same time, clear and cloudless weather can reduce the interference of atmospheric scattering and refraction on spectral data, ensuring that the collected images more truly reflect the actual situation of the wetland in terms of color, brightness, etc.

[0028] The storage and transmission of image data are key links in the entire wetland image collection workflow. When drones are used to collect wetland images, large-capacity, high-speed storage devices should be used. Large-capacity storage devices can ensure that image data can be recorded in real time and completely during long-term flight collection, avoiding data loss due to insufficient storage space. High-speed storage devices can ensure the data writing speed and meet the data storage needs of drones when taking pictures quickly. In order to transmit image data to the data center in a timely manner for subsequent processing and analysis, high-speed wireless communication or satellite communication technology can be used. High-speed wireless communication technology can achieve rapid data return in areas close to the data center and with good signal coverage. It has fast transmission speed and high stability, which can meet the needs of real-time data processing. Satellite communication technology is not limited by geographical distance and terrain. Even in remote wetland areas, the collected image data can be transmitted back to the data center in a timely manner. In the process of data transmission, taking necessary data encryption and security measures is an important means to ensure data security. Encrypting data through encryption algorithms prevents data from being stolen or tampered with during transmission, ensuring the accuracy and security of wetland biological environment monitoring data. At the same time, a data backup mechanism is established to regularly back up image data and store the data in multiple different storage media, such as hard disk arrays, cloud storage, etc., to cope with possible storage device failures or data loss and ensure data integrity and recoverability.

[0029] Figure 1 Flow chart of a method for collaborative online monitoring of a wetland biological environment according to an embodiment of the present application. Figure 2 Schematic diagram of data flow of the wetland biological environment collaborative online monitoring method according to an embodiment of the present application. Figure 1 and Figure 2 As shown, according to the embodiment of the present application, the collaborative online monitoring method of the wetland biological environment includes the following steps: S1, using a color histogram to extract the color features of the wetland image data to obtain a wetland image color feature coding vector; S2, using a directional gradient histogram to extract the HOG features of the wetland image data to obtain a wetland image HOG feature coding vector; S3, performing feature principal component compensation interaction on the wetland image color feature coding vector and the wetland image HOG feature coding vector to obtain a wetland image color-HOG compensated interactive coding feature vector; S4, estimating the vegetation coverage rate based on the wetland image color-HOG compensated interactive coding feature vector to obtain the vegetation coverage.

[0030] In the above-mentioned wetland biological environment collaborative online monitoring method, the step S1 uses a color histogram to extract the color features of the wetland image data to obtain a wetland image color feature coding vector. It should be understood that color is a very intuitive and important feature information in wetland images, which can reflect the inherent properties of the objects to a certain extent, and is relatively less affected by external factors such as light, and has good adaptability to the complex and changeable wetland environment. Since the vegetation, water bodies, soil, etc. in the wetland often have significant differences in color distribution, the present application helps to distinguish different wetland elements by extracting the color features of the wetland image data, and provides an important basis for subsequent vegetation feature analysis and vegetation coverage estimation.

[0031] In a specific example of the present application, the step S1 includes: gridding the wetland image data to obtain a set of wetland image sub-regions; calculating the color histogram of each wetland image sub-region in the set of wetland image sub-regions to obtain a set of wetland image sub-region color histograms; normalizing each wetland image sub-region color histogram in the set of wetland image sub-region color histograms to obtain a set of wetland image sub-region color feature coding vectors; and splicing the set of wetland image sub-region color feature coding vectors to obtain the wetland image color feature coding vector.

[0032] It can also be understood that the color histogram is a method for counting the frequency of occurrence of different colors in an image. It divides the color space of the wetland image into several intervals (i.e., bins), and then counts the number of pixels in each color interval to convert the color information of the wetland image into a one-dimensional numerical vector, i.e., the wetland image color feature coding vector. Each element in the wetland image color feature coding vector corresponds to the frequency of pixels in a specific color interval, thereby being able to fully and accurately describe the color distribution characteristics in the wetland image.

[0033] In the specific implementation, first, the wetland image data is split according to the three color channels of red, green and blue to separate the original color wetland image into these three independent channel images. After the splitting is completed, the pixel counting link is carried out. For each channel, it is necessary to count the number of pixels in each interval one by one according to the pre-set color interval division rules. Taking the red channel as an example, assume that its value range of 0-255 is divided into 16 equal intervals, and each interval spans 16. At this time, it is necessary to traverse each pixel in the red channel image, determine the interval in which its pixel value is located, and accumulate the pixel counts of the corresponding intervals until the number of pixels in each interval is counted.

[0034] When the statistics of the red channel are completed, the same operation is performed on the green channel and the blue channel. The value range of the green channel and the blue channel is also 0-255, and the number of pixels is counted according to the same 16 interval division rule. The statistical results of each channel represent the pixel distribution of the color channel in different color intervals.

[0035] After completing the pixel distribution statistics of the three channels, the statistical results are further arranged and combined in an orderly manner. In this process, the statistical data of the red channel, green channel and blue channel need to be integrated in a certain logical order. For example, the pixel statistical values ​​of the 16 intervals of the red channel are arranged in sequence, followed by the 16 interval statistical values ​​of the green channel, and finally the 16 interval statistical values ​​of the blue channel, thus forming a digital description of the color characteristics of the wetland image.

[0036] In the above-mentioned wetland biological environment collaborative online monitoring method, the step S2 uses the directional gradient histogram to extract the HOG features of the wetland image data to obtain the wetland image HOG feature coding vector. It should be understood that in the wetland image, the morphology, contour and other information of the vegetation also play a key role in the identification of the vegetation, and the directional gradient histogram (HOG) as an effective image feature descriptor can capture the local shape and texture information in the image. Therefore, the present application further uses the directional gradient histogram algorithm to capture the edge gradient direction and intensity information of the vegetation in the wetland image, and then obtains the wetland image HOG feature coding vector. Specifically, the calculation of the wetland image HOG feature is based on the gradient direction and intensity distribution of pixels in the local area of ​​the wetland image. First, the gradient direction and size of each pixel in the wetland image are calculated, and then the image is divided into several small cells, and the number of pixels with different gradient directions in each cell is counted to form a gradient direction histogram. Next, adjacent cells are combined into larger blocks, and the histograms within the blocks are normalized to enhance robustness to factors such as illumination changes. Finally, the normalized histograms of all blocks are sequentially connected to form the wetland image HOG feature encoding vector.

[0037] In a specific example of the present application, the step S2 includes: performing image grayscale and gamma normalization processing on the wetland image data to obtain a wetland grayscale image; calculating the gradients in the horizontal and vertical directions of the wetland grayscale image to obtain a wetland horizontal coordinate direction gradient map and a wetland vertical coordinate direction gradient map; based on the wetland horizontal coordinate direction gradient map and the wetland vertical coordinate direction gradient map, respectively calculating the gradient direction and gradient amplitude of each pixel point in the wetland grayscale image; dividing the wetland grayscale image into local areas to obtain wetland grayscale image blocks. a set of; dividing the gradient direction of each pixel in the wetland grayscale image into 9 gradient direction intervals on the interval [0,π] on average, and calculating the accumulated value of the gradient amplitude of each pixel belonging to each gradient direction interval in each wetland grayscale image block to obtain the directional gradient histogram vector corresponding to each wetland grayscale image block, wherein the length of the directional gradient histogram vector is 9; cascading the directional gradient histogram vectors of each wetland grayscale image block in the set of wetland grayscale image blocks to obtain the wetland image HOG feature coding vector.

[0038] In the above-mentioned wetland biological environment collaborative online monitoring method, the step S3 performs a feature principal component compensation interaction on the wetland image color feature coding vector and the wetland image HOG feature coding vector to obtain the wetland image color-HOG compensation interactive coding feature vector. It should be understood that the wetland image color feature coding vector can reflect the inherent properties of different objects in the wetland image, while the wetland image HOG feature coding vector can capture the morphology and contour information of the vegetation in the wetland image. The two describe the wetland image information from different angles and have a certain complementarity. Therefore, in order to better integrate the multi-dimensional wetland image features and improve the recognition accuracy of wetland vegetation features, the present application introduces a compensatory interactive fusion method based on feature principal component analysis, which first performs principal component analysis on the wetland image color feature coding vector and the wetland image HOG feature coding vector to extract their respective main feature components to reduce redundant information and reduce computational complexity. Then, a compensatory interaction mechanism is further introduced to analyze the feature differences and complementary relationships between the main components of color features and HOG features, and dynamically adjust the weights of the two feature components, so as to achieve information complementarity and fusion between color features and HOG features, and obtain a more comprehensive and accurate description of wetland image features.

[0039] Figure 3 FIG. 1 is a flow chart of step S3 in the wetland biological environment collaborative online monitoring method according to an embodiment of the present application. Figure 3As shown, the step S3 includes: S31, performing feature principal component analysis on the wetland image color feature coding vector and the wetland image HOG feature coding vector to obtain a set of wetland image color feature principal component coding vectors and a set of wetland image HOG feature principal component coding vectors; S32, calculating the wetland image color-HOG feature difference embedding compensation coding weight vector based on the set of wetland image color feature principal component coding vectors and the set of wetland image HOG feature principal component coding vectors; S33, based on the wetland image color-HOG feature difference embedding compensation coding weight vector, performing compensation aggregation interaction on the set of the wetland image color feature principal component coding vector and the set of the wetland image HOG feature principal component coding vector to obtain the wetland image color-HOG compensation interaction coding feature vector.

[0040] Specifically, the step S31 is expressed by the formula:

[0041]

[0042] Among them, V 1 represents the wetland image color feature encoding vector, V 2 represents the HOG feature encoding vector of the wetland image, PCA(·) represents the principal component analysis function, (·) T Represents the transpose of a vector, n is the V 1 and the V 2 The characteristic scale value, C 1 and C 2 Respectively represent the V 1 and the V 2 The covariance matrix, U 1 Indicates that by 1 The matrix composed of the set of principal component coding vectors of wetland image color features obtained by eigenvalue decomposition, Λ 1 Indicates that by 1 The diagonal matrix composed of the set of eigenvalues ​​of the main component of the wetland image color obtained by eigenvalue decomposition, λ 11 ,...,λ 1m represents the color principal component eigenvalues ​​of each wetland image, m is the number of color principal component eigenvalues ​​of the wetland image, and v 11 ,v 12 ,...,v 1m represents each wetland image color feature principal component coding vector in the set of wetland image color feature principal component coding vectors, U 2 Indicates that by 2 The matrix composed of the set of principal component coding vectors of the wetland image HOG features obtained by eigenvalue decomposition, Λ 2 Indicates that by2 The diagonal matrix composed of the set of eigenvalues ​​of the HOG principal component eigenvalues ​​of the wetland image obtained by eigenvalue decomposition, λ 21 ,...,λ 2m Represents the eigenvalues ​​of the HOG principal components of each wetland image, v 21 ,v 22 ,...,v 2m Represents each wetland image HOG feature principal component coding vector in a set of wetland image HOG feature principal component coding vectors.

[0043] That is, in order to extract the most representative characteristic components, the wetland image color characteristic coding vector and the wetland image HOG characteristic coding vector are first subjected to characteristic principal component analysis to extract multiple main characteristic components in the original data, and obtain a set of wetland image color characteristic principal component coding vectors and a set of wetland image HOG characteristic principal component coding vectors. Those skilled in the art should know that characteristic principal component analysis (PCA), as an unsupervised learning method, converts the original data into a set of linearly unrelated variables, namely, principal components, through orthogonal transformation. These principal components are arranged in descending order of variance, thereby retaining the main information in the original data, which helps to more centrally reflect the key information in the wetland image color characteristics and HOG characteristics.

[0044] Specifically, the step S32 includes: first, constructing the set of the wetland image color feature principal component coding vectors and the set of the wetland image HOG feature principal component coding vectors into a wetland image color feature principal component aggregation coding feature map and a wetland image HOG feature principal component aggregation coding feature map, which are expressed as follows:

[0045] F 1 = reshape{v 11 ;v 12 ;…;v 1m}

[0046] F 2 = reshape{v 21 ;v 22 ;…;v 2m}

[0047] Among them, F 1 and F 2 They respectively represent the principal component aggregation coding feature map of wetland image color features and the principal component aggregation coding feature map of wetland image HOG features, and reshape represents feature shape reshaping.

[0048] Here, in order to more intuitively display the spatial distribution information of the wetland image color features and HOG features, the present application further reconstructs the set of the wetland image color feature principal component coding vectors and the set of the wetland image HOG feature principal component coding vectors into feature map forms, respectively, to maintain the spatial structure of the wetland image features, and obtain the wetland image color feature principal component aggregation coding feature map and the wetland image HOG feature principal component aggregation coding feature map, thereby providing a more suitable data structure for subsequent feature interaction analysis.

[0049] Next, the wetland image color feature principal component aggregation coding feature map and the wetland image HOG feature principal component aggregation coding feature map are respectively input into the feature embedding unit to obtain the wetland image color feature weight vector and the wetland image HOG feature weight vector; based on the wetland image color feature weight vector and the wetland image HOG feature weight vector, the wetland image color-HOG feature difference embedding compensation coding weight vector is calculated. Wherein, calculating the wetland image color-HOG feature difference embedding compensation coding weight vector includes: calculating the position difference vector between the wetland image color feature weight vector and the wetland image HOG feature weight vector and taking the absolute value of the position difference vector to obtain the wetland image color-HOG feature difference embedding compensation coding weight vector.

[0050] The above calculation process can be expressed as:

[0051]

[0052] Among them, AvgPool(·) represents the mean pooling function, W 11 and W 12 Respectively represent the F 1 Two learnable weight parameter matrices, σ represents the Sigmoid activation function, W 21 and W 22 Respectively represent the F 2 Two learnable weight parameter matrices, δ represents the normalization function, X 1 ′ represents the wetland image color feature weight vector, X 2 ′ represents the HOG feature weight vector of the wetland image, represents the difference operation, P represents the wetland image color-HOG feature difference embedding compensation encoding weight vector, and |·| represents taking the absolute value.

[0053] That is, by further using a feature embedding unit to perform low-dimensional embedding encoding on the wetland image color feature principal component aggregation coding feature map and the wetland image HOG feature principal component aggregation coding feature map, the relative importance of the wetland image color principal component features and HOG principal component features is quantified through the processing of a multi-layer neural network, and a wetland image color feature weight vector and a wetland image HOG feature weight vector are generated. Then, by performing a differential operation on the wetland image color feature weight vector and the wetland image HOG feature weight vector, the difference and complementarity of color features and HOG features in describing wetland images are revealed, and a wetland image color-HOG feature difference embedding compensation coding weight vector is generated, so as to guide the subsequent feature interaction fusion process to better balance the contribution of color features and HOG features, make up for each other's shortcomings, and more comprehensively describe the vegetation characteristics in the wetland image.

[0054] Specifically, the step S33 includes: respectively calculating the position mean vector of the set of the wetland image color feature principal component coding vector and the set of the wetland image HOG feature principal component coding vector to obtain the wetland image color feature principal component characterization coding vector and the wetland image HOG feature principal component characterization coding vector; based on the wetland image color-HOG feature difference embedding compensation coding weight vector, the wetland image color feature principal component characterization coding vector and the wetland image HOG feature principal component characterization coding vector are aggregated and interacted to obtain the wetland image color-HOG compensation interaction coding feature vector, which is expressed as follows:

[0055]

[0056] v f =σ(C 1×1 (y 1 ⊙P+y 2 ⊙P))

[0057] in, represents the difference operation, P represents the wetland image color-HOG feature difference embedding compensation encoding weight vector, |·| represents taking the absolute value, v 1j represents the jth wetland image color feature principal component coding vector in the set of the wetland image color feature principal component coding vectors, y 1 represents the principal component representation encoding vector of the wetland image color feature, v 2j represents the jth wetland image HOG feature principal component coding vector in the set of the wetland image HOG feature principal component coding vectors, y 2 represents the encoding vector of the principal component representation of the HOG feature of the wetland image, ⊙ represents the dot product, C 1×1 (·) represents the point convolution operation, v fRepresents the wetland image color-HOG compensation interactive encoding feature vector.

[0058] Here, in order to comprehensively consider the color feature principal components and HOG feature principal components of the wetland image in the subsequent multi-dimensional information aggregation interaction, the set of the wetland image color feature principal component coding vector and the set of the wetland image HOG feature principal component coding vector are first integrated by calculating the position mean. Then, the wetland image color-HOG feature difference embedding compensation coding weight vector is used as the adjustment factor to perform weighted fusion on the integrated wetland image color feature principal component and HOG feature principal component, and through point convolution coding and activation function processing, the key information in the fusion feature is further extracted and emphasized to obtain the wetland image color-HOG compensation interaction coding feature vector. In this way, not only the color distribution characteristics of the wetland image and the morphology and contour information of the vegetation are integrated, but also the dynamic adjustment and optimization of the wetland image feature description are realized by introducing the feature compensation interaction mechanism, which helps to more comprehensively and accurately describe the vegetation characteristics in the wetland image and provide more reliable feature input for the subsequent vegetation coverage estimation.

[0059] In the above-mentioned wetland biological environment collaborative online monitoring method, the step S4 estimates the vegetation coverage rate based on the wetland image color-HOG compensation interactive coding feature vector to obtain the vegetation coverage. In a specific example of the present application, the step S4 includes: inputting the wetland image color-HOG compensation interactive coding feature vector into a vegetation coverage rate estimation module based on a decoder to obtain the vegetation coverage. Specifically, the decoder is based on a neural network structure, and gradually restores the spatial resolution of the feature by performing a series of deconvolution, upsampling and other operations on the wetland image color-HOG compensation interactive coding feature vector, and utilizes the nonlinear fitting ability of the neural network to convert the wetland vegetation feature information contained in the wetland image color-HOG compensation interactive coding feature vector into a quantitative estimate of vegetation coverage, thereby achieving accurate prediction of wetland vegetation coverage.

[0060] Furthermore, considering that the wetland image color feature coding vector and the wetland image HOG feature coding vector represent the wetland image color semantic distribution characteristics and the wetland image HOG semantic distribution characteristics respectively, when performing feature principal component compensation interaction, the inherent feature population attribute differences between color features and HOG features will cause fairness differences in the feature principal component compensation levels, thereby affecting the feature distribution interaction inclusiveness of the wetland image color-HOG compensation interaction coding feature vector, and reducing the accuracy of the vegetation coverage rate obtained by its input into the decoder-based vegetation coverage rate estimation module.

[0061] Therefore, preferably, in an embodiment of the present application, when the wetland image color-HOG compensation interactive coding feature vector is input into a decoder-based vegetation coverage estimation module to obtain vegetation coverage, the wetland image color-HOG compensation interactive coding feature vector is first optimized, and the optimization process is:

[0062] Determine the eigenvalue mean and eigenvalue standard deviation corresponding to the wetland image color-HOG compensation interactive encoding eigenvector;

[0063] The first wetland image color-HOG compensation interactive coding fair target vector is obtained by multiplying the dot-subtracted vector of the wetland image color-HOG compensation interactive coding feature vector and the eigenvalue mean with the eigenvalue standard deviation, and the second wetland image color-HOG compensation interactive coding fair target vector is obtained by multiplying the dot-subtracted vector of the wetland image color-HOG compensation interactive coding feature vector and the eigenvalue standard deviation with the eigenvalue mean;

[0064] After performing a dot multiplication of the bit-by-bit reciprocal of the second wetland image color-HOG compensation interactive coding fair target vector and the first wetland image color-HOG compensation interactive coding fair target vector, a bit-by-bit logarithm with a base of 2 is calculated to obtain a wetland image color-HOG compensation interactive coding information correction vector; and

[0065] The square root of the quotient of the eigenvalue mean divided by the eigenvalue standard deviation is multiplied by the weight hyperparameter, and then added to the wetland image color-HOG compensation interactive coding information correction vector point to obtain an optimized wetland image color-HOG compensation interactive coding eigenvector.

[0066] Therefore, taking into account the attribute level fairness differences of the feature populations corresponding to the fusion features of the wetland image color-HOG compensation interactive coding feature vector, in order to improve the interactive inclusiveness of the wetland image color-HOG compensation interactive coding feature vector under the feature distribution diversity, the crossover probability value constraint based on the wetland image color-HOG compensation interactive coding feature vector is used as the interactive fairness target representation to correct the interactive propagation of the group feature information of the wetland image color-HOG compensation interactive coding feature vector, and the unified statistical feature response interaction based on the wetland image color-HOG compensation interactive coding feature vector is used as the feature distribution multi-level fairness target bias to achieve the robust distribution fairness unified representation of the wetland image color-HOG compensation interactive coding feature vector, forming a fair collaboration paradigm under the feature distribution framework of the wetland image color-HOG compensation interactive coding feature vector, and improving the accuracy of the vegetation coverage rate obtained by the vegetation coverage rate estimation module based on the decoder.

[0067] After obtaining the vegetation coverage, it can be associated with environmental factor data such as temperature, salinity, water level, and pH value for response analysis, so as to deeply explore the complex relationship between wetland vegetation and environmental factors and understand the dynamic changes of wetland ecosystems. For example, taking salinity as an environmental factor, compare the vegetation coverage parameters under different salinity conditions. When the salinity changes, observe whether the vegetation coverage changes accordingly, as well as the corresponding changes in vegetation density and plant height. If the salinity increases, the growth of some wetland vegetation that is sensitive to salinity may be inhibited, which is manifested as a decrease in vegetation coverage, density, and plant height. By establishing a mathematical model, such as a linear regression model or a nonlinear regression model, the relationship between salinity and vegetation coverage is quantitatively analyzed, and the response function between the two is determined. In addition to salinity, similar analysis methods can be used for other environmental factors such as temperature, water level, and pH value to explore their inherent connection with various parameters of wetland vegetation.

[0068] In the stage of interpretation and application of results, the status of wetland ecosystems is evaluated based on the response relationship between wetland vegetation and changes in environmental factors obtained through data analysis. If it is found that changes in certain environmental factors lead to adverse responses in wetland vegetation, such as a continuous decline in vegetation coverage and a reduction in species diversity, appropriate protection and management measures need to be taken in a timely manner. For example, if the analysis results show that excessive fluctuations in water levels are the main cause of wetland vegetation degradation, water conservancy facilities such as dams and sluices can be built to regulate water levels, maintain relative stability of water levels, and create a suitable growth environment for wetland vegetation.

[0069] In summary, the wetland biological environment collaborative online monitoring method according to the embodiment of the present application is explained, which uses color histogram and directional gradient histogram to extract color features and HOG features of wetland image data respectively, and introduces deep learning algorithms to perform principal component analysis and compensatory interactive fusion between cross-domain features on the color features and HOG features of wetland images, so as to achieve accurate recognition of vegetation features in wetland images and intelligent estimation of vegetation coverage, and then conduct correlation response analysis between wetland organisms and environmental factors on this basis. In this way, the recognition accuracy of wetland vegetation features can be improved, and the adaptability of the model to different lighting conditions and complex backgrounds can be enhanced, thereby providing more reliable data support for wetland protection and management.

[0070] Furthermore, the present application also provides a wetland biological environment collaborative online monitoring system.

[0071] Figure 4 FIG. 1 is a block diagram of a wetland biological environment collaborative online monitoring system according to an embodiment of the present application. Figure 4As shown, according to the wetland biological environment collaborative online monitoring system 100 of the embodiment of the present application, it includes: a wetland image color feature extraction module 110, which is used to extract the color features of the wetland image data using a color histogram to obtain a wetland image color feature coding vector; a wetland image HOG feature extraction module 120, which is used to extract the HOG features of the wetland image data using a directional gradient histogram to obtain a wetland image HOG feature coding vector; a principal component compensation interaction module 130, which is used to perform feature principal component compensation interaction on the wetland image color feature coding vector and the wetland image HOG feature coding vector to obtain a wetland image color-HOG compensation interaction coding feature vector; a vegetation coverage estimation module 140, which is used to estimate the vegetation coverage based on the wetland image color-HOG compensation interaction coding feature vector to obtain vegetation coverage.

[0072] Here, those skilled in the art can understand that the specific operations of each module in the above-mentioned wetland biological environment collaborative online monitoring system have been referred to above. Figures 1 to 3 The wetland biological environment collaborative online monitoring method has been introduced in detail, and therefore, its repeated description will be omitted.

[0073] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit and scope of the technical solution of the present invention.

Claims

1. A wetland biological environment collaborative online monitoring method, comprising: Synchronously acquiring wetland image data and environmental element data of the wetland; and calculating vegetation coverage based on the wetland image data; Based on the environmental factor data and the vegetation coverage, the response of wetland vegetation to the change of environmental factors is obtained, characterized in that the vegetation coverage is calculated based on the wetland image data, including: Extracting the color features of the wetland image data using a color histogram to obtain a wetland image color feature encoding vector; Using a histogram of directional gradients to extract the HOG features of the wetland image data to obtain a wetland image HOG feature encoding vector; Performing a principal component compensation interaction on the wetland image color feature coding vector and the wetland image HOG feature coding vector to obtain a wetland image color-HOG compensation interaction coding feature vector; The vegetation coverage rate is estimated based on the wetland image color-HOG compensation interactive coding feature vector to obtain the vegetation coverage.

2. The wetland biological environment collaborative online monitoring method according to claim 1 is characterized in that: The color feature of the wetland image data is extracted using a color histogram to obtain a wetland image color feature encoding vector, including: Gridding the wetland image data to obtain a set of wetland image sub-areas; Calculating the color histogram of each wetland image sub-region in the set of wetland image sub-regions to obtain a set of wetland image sub-region color histograms; Normalizing each wetland image sub-region color histogram in the set of wetland image sub-region color histograms to obtain a set of wetland image sub-region color feature coding vectors; The set of color feature coding vectors of the wetland image sub-regions are concatenated to obtain the color feature coding vector of the wetland image.

3. The wetland biological environment collaborative online monitoring method according to claim 2 is characterized in that: The HOG feature of the wetland image data is extracted using a histogram of directional gradients to obtain a wetland image HOG feature encoding vector, including: Performing image grayscale conversion and gamma normalization processing on the wetland image data to obtain a wetland grayscale image; Calculating the gradients in the horizontal and vertical directions of the wetland grayscale image to obtain a wetland horizontal direction gradient map and a wetland vertical direction gradient map; Based on the wetland horizontal coordinate direction gradient map and the wetland vertical coordinate direction gradient map, respectively calculating the gradient direction and gradient amplitude of each pixel point in the wetland grayscale image; Dividing the wetland grayscale image into local areas to obtain a set of wetland grayscale image blocks; The gradient direction of each pixel in the wetland grayscale image is evenly divided into 9 gradient direction intervals in the interval [0,π], and the cumulative value of the gradient amplitude of each pixel belonging to each gradient direction interval in each wetland grayscale image block is calculated respectively to obtain the directional gradient histogram vector corresponding to each wetland grayscale image block, wherein the length of the directional gradient histogram vector is 9; The directional gradient histogram vectors of each wetland grayscale image block in the set of wetland grayscale image blocks are cascaded to obtain the wetland image HOG feature coding vector.

4. The wetland biological environment collaborative online monitoring method according to claim 3 is characterized in that: Performing a principal component compensation interaction on the wetland image color feature coding vector and the wetland image HOG feature coding vector to obtain a wetland image color-HOG compensation interaction coding feature vector, including: Performing feature principal component analysis on the wetland image color feature coding vector and the wetland image HOG feature coding vector to obtain a set of wetland image color feature principal component coding vectors and a set of wetland image HOG feature principal component coding vectors; Calculate the wetland image color-HOG feature difference embedding compensation coding weight vector based on the set of the wetland image color feature principal component coding vectors and the set of the wetland image HOG feature principal component coding vectors; Based on the wetland image color-HOG feature difference embedding compensation coding weight vector, the set of the wetland image color feature principal component coding vectors and the set of the wetland image HOG feature principal component coding vectors are compensated and aggregated to obtain the wetland image color-HOG compensation interaction coding feature vector.

5. The wetland biological environment collaborative online monitoring method according to claim 4 is characterized in that: Based on the set of the wetland image color feature principal component coding vectors and the set of the wetland image HOG feature principal component coding vectors, a wetland image color-HOG feature difference embedding compensation coding weight vector is calculated, including: The set of the wetland image color feature principal component coding vectors and the set of the wetland image HOG feature principal component coding vectors are constructed into a wetland image color feature principal component aggregation coding feature map and a wetland image HOG feature principal component aggregation coding feature map; Inputting the wetland image color feature principal component aggregation coding feature map and the wetland image HOG feature principal component aggregation coding feature map into a feature embedding unit to obtain a wetland image color feature weight vector and a wetland image HOG feature weight vector respectively; Based on the wetland image color feature weight vector and the wetland image HOG feature weight vector, the wetland image color-HOG feature difference embedding compensation coding weight vector is calculated.

6. The wetland biological environment collaborative online monitoring method according to claim 5 is characterized in that: Calculating the wetland image color-HOG feature difference embedding compensation coding weight vector based on the wetland image color feature weight vector and the wetland image HOG feature weight vector, including: The position difference vector between the wetland image color feature weight vector and the wetland image HOG feature weight vector is calculated, and the absolute value of the position difference vector is taken to obtain the wetland image color-HOG feature difference embedding compensation coding weight vector.

7. The wetland biological environment collaborative online monitoring method according to claim 6 is characterized in that: Based on the wetland image color-HOG feature difference embedding compensation coding weight vector, the set of the wetland image color feature principal component coding vector and the set of the wetland image HOG feature principal component coding vector are compensated and aggregated to obtain the wetland image color-HOG compensation interaction coding feature vector, including: Respectively calculating the position mean vectors of the set of the wetland image color feature principal component coding vectors and the set of the wetland image HOG feature principal component coding vectors to obtain the wetland image color feature principal component representation coding vector and the wetland image HOG feature principal component representation coding vector; Based on the wetland image color-HOG feature difference embedding compensation coding weight vector, the wetland image color feature principal component representation coding vector and the wetland image HOG feature principal component representation coding vector are aggregated and interacted to obtain the wetland image color-HOG compensation interaction coding feature vector.

8. The wetland biological environment collaborative online monitoring method according to claim 7, characterized in that: Estimating vegetation coverage based on the wetland image color-HOG compensation interactive coding feature vector to obtain the vegetation coverage includes: The wetland image color-HOG compensation interactive coding feature vector is input into a decoder-based vegetation coverage estimation module to obtain the vegetation coverage.

9. A wetland biological environment collaborative online monitoring system, characterized in that: include: A wetland image color feature extraction module is used to extract the color features of the wetland image data using a color histogram to obtain a wetland image color feature encoding vector; A wetland image HOG feature extraction module is used to extract the HOG features of the wetland image data using a directional gradient histogram to obtain a wetland image HOG feature encoding vector; A principal component compensation interaction module is used to perform principal component compensation interaction on the wetland image color feature coding vector and the wetland image HOG feature coding vector to obtain a wetland image color-HOG compensation interaction coding feature vector; The vegetation coverage estimation module is used to estimate the vegetation coverage based on the wetland image color-HOG compensation interactive coding feature vector to obtain the vegetation coverage.

Images