Landscape landscaping monitoring method based on image recognition

Abstract

The invention relates to the technical field of environment monitoring, in particular to a landscape landscaping monitoring method based on image recognition, and the method comprises the steps: obtaining data; determining a suspicious viewing point; determining a sight management and control type viewing point and a space creation type viewing point; determining an abnormal viewing point; determining a target viewing point; and adjusting a threshold value and carrying out early warning. According to the method, multiple key parameters of an ornamental point image are acquired, preliminary screening is performed by fusing green vision, colors and transparency, then landscape types are distinguished by analyzing fluctuation correlation between the transparency and the prominence, then anomaly traceability is performed on different types by adopting special indexes, and furthermore, the landscape types are subjected to abnormal traceability. According to the method, clustering analysis is carried out in combination with spatial positions and reasons of abnormal points, isolated points or regional events are identified, finally, a judgment threshold value is dynamically adjusted by utilizing data distribution of target viewing points, and the problems of low abnormal identification accuracy and risk early warning lag caused by excessive dependence on static data are effectively solved.

Description

Technical Field

[0001] This invention relates to the field of environmental monitoring technology, and in particular to a landscape greening monitoring method based on image recognition. Background Technology

[0002] With the deepening of urbanization and residents' pursuit of a high-quality living environment, landscape architecture, as an important component of the urban ecosystem and a vital public leisure space, is continuously expanding in scale, making the demand for refined and intelligent maintenance and management increasingly urgent. Traditional landscape greening monitoring relies heavily on regular manual inspections, which is not only inefficient and costly, but also has highly subjective evaluation standards, making it difficult to achieve large-scale, real-time dynamic monitoring and early warning. Furthermore, it cannot identify and trace the root causes of complex problems such as latent pests and diseases, growth stress, and landscape aesthetic degradation in their early stages.

[0003] Chinese Patent Application Publication No. CN120472321A discloses a garden greening monitoring system and method. The method includes: an acquisition module, an analysis module, an identification and selection module, a weighted fusion module, and an early warning module. The acquisition module acquires the following image data in stages: first-level images, remote sensing images acquired from above the garden via satellite or UAV; second-level images, selectively acquired combined images of the garden from below, images of fallen leaves from the front or back, or images of fallen leaves from the back, based on correlation determination results. The analysis module extracts vegetation indices, disease targets, leaf health characteristics, and fallen leaf status characteristics. The identification and selection module performs the following: calculating the vegetation index based on the first-level images to identify preliminary risk areas; analyzing the correlation anomaly indicators of different image data in the preliminary risk areas, including the deviation rate of the vegetation index from historical data and the matching degree of heat radiation and chlorophyll content; generating a dynamic acquisition strategy based on the correlation anomaly indicators to trigger the acquisition module to acquire second-level images in a targeted manner; the weighted fusion module adaptively weights and fuses multi-source features to calculate the comprehensive wilting risk level; and the early warning module generates early warning signals and early warning levels.

[0004] Therefore, the landscape greening monitoring system has the following problems: it relies on accurate spatiotemporal registration of multi-source data, and it is difficult and costly to achieve pixel-level high-precision matching. Furthermore, any registration error will directly lead to the failure of correlation determination. In real and variable environments, the indicative nature of different features will vary with the season, tree species, and disease type. Fixed or simple rule weighting strategies are difficult to guarantee the accuracy of the assessment. Summary of the Invention

[0005] To address this, the present invention provides a landscape greening monitoring method based on image recognition, which overcomes the problems of low accuracy in anomaly identification and delayed risk warning caused by over-reliance on static data in existing technologies through multi-dimensional data analysis and dynamic adjustment mechanisms.

[0006] To achieve the above objectives, the present invention provides a landscape greening monitoring method based on image recognition, comprising: Real-time acquisition of greening images of various viewing points in the landscape garden, and extraction of greening rate, color vitality value, image transparency, morphological regularity, vegetation encroachment rate of the visual corridor, prominence of visual focus and visibility from the greening images. The landscape index is determined based on the green visibility rate, the color vitality value, and the image transparency, and several suspicious viewing points are identified based on the comparison results between the landscape index and the preset index threshold. Based on the temporal variation characteristics of the image transparency and prominence of each of the suspected viewing points, a type tendency value is determined, and based on the comparison results of the type tendency value and the preset tendency threshold, a number of view control type viewing points and a number of space creation type viewing points are determined. Based on the aforementioned viewpoints with controlled line of sight, the cause of the anomaly is determined by the line of sight obstruction index calculated according to the occupancy rate and the visibility, thereby obtaining several abnormal viewpoints. Based on the aforementioned spatial creation viewing points, the causes of anomalies are determined by the spatial quality index calculated according to the color vitality value and the morphological regularity, thereby obtaining several abnormal viewing points. Based on the spatial location of each abnormal viewing point and the cause of the abnormality, several target viewing points are determined; The preset index threshold is adjusted according to the numerical distribution characteristics of the landscape index of each target viewing point, and the preset tendency threshold is adjusted according to the numerical distribution characteristics of the type tendency value of each target viewing point; Greening intervention early warning is issued for all target viewpoints obtained after adjusting the preset index threshold and the preset tendency threshold.

[0007] Furthermore, the process of determining a landscape index based on the green visibility rate, the color vitality value, and the image transparency, and identifying several suspicious viewing points based on the comparison between the landscape index and a preset index threshold, includes: The landscape index is calculated based on the green visibility rate, the color vitality value, and the image transparency. When the landscape index is less than the preset index threshold, the viewing point to be tested is determined as a suspicious viewing point, thereby identifying several suspicious viewing points.

[0008] Furthermore, the process of determining the type tendency value based on the temporal variation characteristics of the image clarity and prominence of each of the aforementioned suspicious viewing points includes: Calculate several transparency fluctuation values ​​based on the image transparency within a preset judgment time period; Calculate several prominence fluctuation values ​​based on the prominence within the preset determination time period; The type tendency value is calculated based on all the transparency fluctuation values ​​and all the prominence fluctuation values.

[0009] Furthermore, the process of determining several viewpoints for line-of-sight control and several viewpoints for spatial creation based on the comparison results of type tendency values ​​and preset tendency thresholds includes: When the type tendency value is greater than or equal to the preset tendency threshold, the suspicious viewing point is determined to be the line-of-sight control type viewing point; when the type tendency value is less than the preset tendency threshold, the suspicious viewing point is determined to be the space creation type viewing point.

[0010] Furthermore, the process of determining the cause of the anomaly based on the occupancy rate and the visibility-calculated visual obstruction index, thereby obtaining several abnormal viewing points, includes: Calculate the rate of change of occupancy rate based on the occupancy rate within a preset time period; Calculate the visibility change rate based on the visibility within the preset time period; Calculate the line obstruction index based on the occupancy rate change rate and the visibility change rate; Based on the comparison between the line-of-sight obstruction index and the preset obstruction threshold, the abnormality of the line-of-sight control viewing point is determined to be plant encroachment, thereby identifying the line-of-sight control viewing point as the abnormal viewing point and obtaining several abnormal viewing points.

[0011] Furthermore, the process of determining the cause of anomalies based on the spatial quality index calculated from the color vitality value and the morphological regularity, thereby obtaining several abnormal viewing points, includes: Two-dimensional data points are constructed based on the color vibrancy value and the shape regularity. Calculate the health distance based on the two-dimensional data points and the preset health benchmark points; Based on the comparison between the healthy distance and the preset healthy distance threshold, the abnormality of the spatially designed viewing point is determined to be plant disqualification, thereby identifying the spatially designed viewing point as the abnormal viewing point and obtaining several abnormal viewing points.

[0012] Furthermore, the process of determining several target viewing points based on the spatial location of each of the aforementioned abnormal viewing points and the cause of the abnormality includes: Calculate the abnormal distance based on the position coordinates of any two abnormal viewing points, and mark the two abnormal viewing points as the first abnormal viewing point and the second abnormal viewing point, respectively. When the abnormal distance is greater than or equal to a preset abnormal distance threshold, the first abnormal viewing point and the second abnormal viewing point are determined as the target viewing point, so as to determine a number of target viewing points; When the abnormal distance is less than the preset abnormal distance threshold, several target viewing points are determined based on the abnormal cause of the first abnormal viewing point and the abnormal cause of the second abnormal viewing point.

[0013] Furthermore, the process of determining several target viewing points based on the cause of the anomaly of the first abnormal viewing point and the cause of the anomaly of the second abnormal viewing point includes: Based on the first comparison result of the abnormal cause of the first abnormal viewing point and the abnormal cause of the second abnormal viewing point, the first abnormal viewing point and the second abnormal viewing point are determined as the target viewing points, thereby identifying a number of target viewing points; Based on the second comparison result of the abnormal cause of the first abnormal viewing point and the abnormal cause of the second abnormal viewing point, an abnormal event in the area is determined; Based on the regional anomaly events, the regional center is determined according to the first and second anomaly viewing points. An abnormal region is determined based on the region center and a preset abnormal radius. All abnormal viewing points within the abnormal region are identified as target viewing points to determine a number of target viewing points. The landscape index of all the viewing points to be tested within the abnormal region is extracted. The rate of change of each landscape index is calculated. When the rate of change of the index is greater than a preset landscape change threshold, the viewing point to be tested is identified as the target viewing point to determine a number of target viewing points.

[0014] Furthermore, the process of adjusting the preset index threshold based on the numerical distribution characteristics of the landscape index at each target viewing point includes: Calculate the upper quartiles based on all the landscape indices mentioned; The preset exponential threshold is adjusted based on the comparison result between the upper quartile and the preset exponential threshold.

[0015] Furthermore, the process of adjusting the preset tendency threshold based on the numerical distribution characteristics of the type tendency values ​​of each target viewing point includes: Clustering is performed on all the aforementioned type tendency values ​​to obtain a first candidate cluster and a second candidate cluster; Calculate the median of all the type tendency values ​​within the first candidate cluster to obtain the first typical tendency value; Calculate the median of all the type tendency values ​​within the second candidate cluster to obtain the second typical tendency value; A classification reference value is calculated based on the first typical tendency value and the second typical tendency value. A tendency deviation is calculated based on the classification reference value and a preset tendency threshold. The preset tendency threshold is adjusted based on the comparison between the tendency deviation and the preset tendency deviation threshold.

[0016] Compared with existing technologies, the beneficial effects of this invention are as follows: by collecting multiple key parameters of viewing point images, and using a landscape index calculated by integrating green vision, color, and transparency for preliminary aesthetic screening, the invention further analyzes the type tendency value obtained based on the fluctuation correlation between transparency and prominence to distinguish whether the root cause of the problem belongs to the category of sightline control or spatial creation. Based on the principles of spatial vision and landscape aesthetics theory, it ensures that the classification is aligned with the goals of professional garden management. Then, for different types, it uses specific quantitative indices, namely the sightline obstruction index and the spatial quality health distance, to trace the source of anomalies, achieving data-driven diagnosis from phenomenon to cause. Furthermore, it combines the spatial location of anomalies with the consistency of causes to perform cluster analysis, identifying isolated point problems or potential regional events. Finally, it uses the data distribution of target viewing points to dynamically adjust the judgment threshold, enabling the entire system to have continuous learning and optimization capabilities, forming a complete closed loop. This significantly improves the scientific, forward-looking, and refined level of maintenance management, effectively solving the problems of low accuracy in anomaly identification and delayed risk warning caused by over-reliance on static data.

[0017] Furthermore, by constructing a comprehensive evaluation model integrating ecology, aesthetics, and spatial perception, an objective diagnosis of landscape aesthetics was achieved. Green view rate directly reflects the ecological foundation and visual green volume of vegetation cover; color vitality value characterizes the color richness and visual appeal of the landscape; and image transparency relates to the openness and visual comfort of the space. These three factors define the visual health status of the landscape from the dimensions of green volume, color, and space, respectively. Normalization eliminates the dimensional differences in the original data, and pre-set weights are introduced for weighted summation, enabling the correlation and fusion of multi-dimensional heterogeneous data. The resulting landscape index becomes a unified and comparable benchmark. By comparing it with preset index thresholds, suspicious viewing points that fail to meet the standards in the comprehensive dimensions can be accurately identified, laying a reliable data foundation and avoiding misjudgments caused by the one-sidedness of a single indicator.

[0018] Furthermore, by analyzing the coordinated fluctuation patterns of landscape visual features over time, intelligent prediction of landscape types was achieved. The fluctuations in image transparency and visual focus prominence respectively characterize the temporal stability of spatial openness and visual guidance. In ideal visual control areas, due to strong constraints from human planning, their fluctuations should show a high positive correlation. However, in spatial creation areas dominated by natural changes, the correlation between the two fluctuations is weaker. By normalizing the fluctuation sequence and calculating the Pearson correlation coefficient, this step quantifies the coordinated relationship between the two fluctuations over time into a scalar value, namely the type tendency value. Essentially, this constructs a mathematical indicator to measure the intensity of visual control, enabling a deep logical connection between static space and visual indicators at the dynamic data level. This allows for the automatic identification of landscape types that conform to management intentions from the data, providing an accurate classification basis for subsequent targeted anomaly diagnosis.

[0019] Furthermore, by intelligently mapping continuous data reflecting the intensity of visual control to discrete landscape categories with clear management semantics, a higher type tendency value indicates a highly coordinated fluctuation in transparency and prominence, revealing that landscape visual elements are in a controlled and interconnected state of change. This aligns with the professional characteristics of visual control areas, which require strict management of visual corridors and focal sequences. Conversely, a lower type tendency value indicates that the fluctuations of the two are relatively independent, reflecting a more natural and random landscape change, which fits the management goal of space creation areas, which pursues a diverse and flexible visual experience. The preset tendency threshold here acts as a classification decision boundary, enabling the system to automatically and consistently transform data features into type labels that can directly guide differentiated maintenance decisions. Thus, after a problem is discovered, it can accurately pinpoint the type of problem and determine the appropriate strategy for handling it, achieving a deep correlation and closed loop between data flow and maintenance business flow.

[0020] Furthermore, based on the principle that the decrease in visibility and the increase in vegetation encroachment rate should show a significant negative correlation over time in the visual control area, the rate of change of the two is calculated and the negative of the ratio is obtained. This physical relationship is creatively transformed into a visual obstruction index: when plant growth does indeed constitute obstruction, the index will show a large positive value because the rate of decrease in visibility is negative and the rate of increase in encroachment rate is positive; conversely, if the change in visibility is caused by other factors, the index will not rise abnormally. This establishes a strong logical connection between the trend of change in the data layer and the abnormal cause (plant encroachment) in the physical layer. The preset obstruction threshold is used as an empirical critical point, enabling the system to automatically screen out those points where the visibility deteriorates significantly due to the dynamic growth of vegetation. Thus, general problems in visual control are accurately located as abnormal viewing points that require intervention, realizing an intelligent leap from type judgment to cause diagnosis.

[0021] Furthermore, by transforming subjective judgments of plant landscape disqualification into objective spatial distance measurements, color vitality value and morphological regularity define the two core aesthetic dimensions of plant visual health: color and form. Together, they constitute a two-dimensional evaluation plane. After normalization to eliminate dimensions, the state of each viewing point is mapped to a data point on the plane. The preset health benchmark represents the ideal state of excellent color and form in the context of the garden. Calculating the Euclidean distance between the two essentially measures the comprehensive deviation between the current state and the ideal state, thus mathematically linking the two originally separate indicators. The geometric distance intuitively reflects the degree of decline in comprehensive aesthetic quality. When the distance exceeds the preset threshold, it indicates an unacceptable deviation in both color and form. The system can then accurately diagnose the plant as disqualified, thereby transforming the vague unattractive problem in the spatial creation area into a clear, measurable, and intervention-required specific abnormal cause.

[0022] Furthermore, by introducing spatial correlation analysis, discrete abnormal alarm points are upgraded to problem areas with management significance. By calculating the Euclidean distance between abnormal points and comparing it with a preset threshold, the spatial density and diffusion probability of abnormal events are first quantified at the data level. When the distance between two points is less than the threshold, it means that they are highly correlated in physical space. At this time, further comparison of the causes of the anomalies becomes the key basis for decision-making, enabling the system to intelligently distinguish between sporadic cases and potential regional risks. This optimizes maintenance decisions from responding to single-point anomalies to proactively identifying and intervening in problem patterns, realizing data-driven, precise and efficient resource scheduling and early warning upgrades.

[0023] Furthermore, by integrating three types of data—spatial location, cause of anomalies, and temporal changes—a dynamic risk assessment model was constructed. When adjacent anomaly points share the same cause, they are identified as events originating from the same region, and their potential impact range is delineated using their geometric center and a preset radius. This aligns with the objective law of the continuous spatial spread of problems such as pests and diseases and improper maintenance. Moreover, the system not only monitors the anomaly points already discovered within the range but also all points to be monitored. By calculating the rate of change of their landscape index and comparing it with a threshold, it essentially monitors whether healthy points within the monitoring area show an early trend of evolving into anomalies. This creates a deep correlation between spatial clustering information, problem attribute information, and temporal early warning signals, enabling the system to go beyond passively recording existing problems and proactively capture the vanguard of potential spread. This shifts the determination of target viewing points from dealing with existing anomalies to intercepting developing risks, achieving optimal allocation and forward-looking deployment of maintenance resources.

[0024] Furthermore, by calculating the average landscape index of the selected target viewing points, a data benchmark reflecting the typical level of the current substandard group is obtained. The relative deviation of this average value from the current standard is calculated, which essentially quantifies the gap between common problems and management requirements. When this deviation exceeds the sensitivity threshold, a threshold adjustment based on a preset coefficient is triggered. This is because if the index of a large number of problem points is concentrated at a level significantly lower than the current threshold, it may indicate that the existing standard has deviated from the generally achievable state of the landscape, or that the overall maintenance level has undergone a phased change. By automatically and slightly lowering the threshold, the system can avoid continuously generating a large number of invalid alarms due to the rigidity of the standard, and keep the monitoring focus in sync with the achievable and generally healthy state of the garden. This makes the anomaly judgment more practically instructive and provides a self-calibrating reference benchmark for the dynamic allocation of maintenance resources.

[0025] Furthermore, by performing unsupervised clustering on the type tendency values ​​of the identified outliers, the natural distribution centers of the two types of problems—sight control and space creation—in the current data are discovered, namely the first and second typical tendency values. The weighted average of these two center values ​​(classification reference value) is used as the theoretically optimal classification boundary suggested by the current data. By calculating the relative deviation between this theoretical boundary and the current application boundary (preset tendency threshold), the degree of mismatch between the real-world problem pattern and the existing classification rules is quantified. When the absolute value of the deviation exceeds the sensitivity threshold, the threshold is fine-tuned to ensure that the classification rules continuously approach the optimal solution revealed by the data. This establishes a strong correlation between the abstract landscape type determination and the actual distribution of specific, dynamically changing outliers, achieving closed-loop self-optimization of the classifier performance in application. This ensures that the system can continuously adapt to the long-term impacts of plant growth, seasonal changes, and shifts in maintenance priorities. Attached Figure Description

[0026] Figure 1 This is a flowchart of the landscape greening monitoring method based on image recognition in this embodiment; Figure 2 This embodiment defines the logic diagram for determining suspicious viewing points. Figure 3 This is a logic diagram for determining the landscape type in this embodiment; Figure 4 This embodiment uses a logic diagram to determine the cause of the anomaly and obtain the abnormal viewing point. Detailed Implementation

[0027] To make the objectives and advantages of the present invention clearer, the present invention will be further described below with reference to embodiments; it should be understood that the specific embodiments described herein are merely for explaining the present invention and are not intended to limit the present invention.

[0028] Preferred embodiments of the present invention will now be described with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are merely illustrative of the technical principles of the present invention and are not intended to limit the scope of protection of the present invention.

[0029] Please see Figure 1 The diagram shown is a flowchart of the landscape greening monitoring method based on image recognition in this embodiment. This embodiment provides a landscape greening monitoring method based on image recognition, including: Real-time acquisition of greening images of various viewing points in the landscape garden, and extraction of greening rate, color vitality value, image transparency, morphological regularity, vegetation encroachment rate of the visual corridor, prominence of visual focus and visibility from the greening images. The landscape index is determined based on the green visibility rate, the color vitality value, and the image transparency, and several suspicious viewing points are identified based on the comparison results between the landscape index and the preset index threshold. Based on the temporal variation characteristics of the image transparency and prominence of each of the suspected viewing points, a type tendency value is determined, and based on the comparison results of the type tendency value and the preset tendency threshold, a number of view control type viewing points and a number of space creation type viewing points are determined. Based on the aforementioned viewpoints with controlled line of sight, the cause of the anomaly is determined by the line of sight obstruction index calculated according to the occupancy rate and the visibility, thereby obtaining several abnormal viewpoints. Based on the aforementioned spatial creation viewing points, the causes of anomalies are determined by the spatial quality index calculated according to the color vitality value and the morphological regularity, thereby obtaining several abnormal viewing points. Based on the spatial location of each abnormal viewing point and the cause of the abnormality, several target viewing points are determined; The preset index threshold is adjusted according to the numerical distribution characteristics of the landscape index of each target viewing point, and the preset tendency threshold is adjusted according to the numerical distribution characteristics of the type tendency value of each target viewing point; Greening intervention early warning is issued for all target viewpoints obtained after adjusting the preset index threshold and the preset tendency threshold.

[0030] This embodiment acquires real-time greening images using image acquisition devices (such as cameras) deployed at various viewing points in the landscape garden. Then, computer vision and image processing technologies are used to automatically quantify and extract seven core feature indicators from the images. Green view rate refers to the percentage of area occupied by vegetation (mainly green plants) pixels in the image, a core indicator for measuring the visual quantity of green coverage, obtained through color space segmentation (such as the green range threshold in HSV space) combined with a semantic segmentation model; color vitality value is a comprehensive evaluation value reflecting the richness, saturation, and harmony of landscape colors, quantified by analyzing the color difference, saturation variance, and hue distribution of the image in color spaces such as Lab; image transparency is an indicator for evaluating the sense of spatial openness and visual depth, related to the amount of obstruction, assessed through depth estimation and scene segmentation; morphological regularity is an indicator that quantifies the plant outline, the neatness of the pruning, and the geometric order, obtained through edge detection and outline extraction. The degree of matching with standard geometry is calculated; the occupancy rate of vegetation in the visual corridor specifically refers to the ratio of vegetation pixels occupying the preset visual space within the defined visual corridor area, which is obtained by calculating the area ratio after delineating the region of interest (ROI) in the image and performing vegetation semantic segmentation; the prominence of the visual focus is an indicator that evaluates the degree to which preset or automatically identified visual focuses (such as sculptures or distinctive trees) attract attention in the image, which is obtained by calculating the saliency value of the focus area through a visual saliency detection model (such as a frequency-tuned algorithm); visibility is an indicator that characterizes the clarity of distant objects under atmospheric conditions or slight occlusion, which can usually be indirectly reflected by an image haze concentration detection model or by calculating local contrast and clarity.

[0031] The preset index threshold is a critical value used to determine whether the landscape index is qualified. It depends on the landscape index of historically healthy viewing points and the results of expert evaluation, and is usually set between 0.6 and 0.8. In this embodiment, it is set to 0.7, which can effectively filter out viewing points with obviously poor overall performance, and will not generate too many invalid alarms due to an excessively high standard, thus ensuring the processing efficiency of subsequent diagnostic procedures.

[0032] The preset tendency threshold is a classification decision boundary value used to divide continuous type tendency values ​​into two landscape types: line-of-sight control and spatial creation. It depends on the definition criteria for the visual control intensity of the two landscape types and the actual distribution of the tendency values ​​of the two types of samples in historical data, and is usually set between 0 and 0.6. In this embodiment, it is set to 0.4, requiring a relatively clear positive correlation between fluctuations in image transparency and salience to determine it as a line-of-sight control type requiring strict management, thus improving the accuracy of the classification.

[0033] By collecting multiple key parameters from images of viewing points, and using a landscape index calculated by integrating green vision, color, and transparency for initial screening, the system further analyzes the type tendency values ​​obtained from the fluctuation correlation between transparency and prominence to distinguish whether the root cause of the problem belongs to the category of sightline control or spatial creation. Based on the principles of spatial vision and landscape aesthetics theory, the classification is ensured to align with the goals of professional garden management. Next, specific quantitative indices, namely the sightline obstruction index and the spatial quality health distance, are used for anomaly tracing for different types, achieving data-driven diagnosis from phenomenon to cause. Furthermore, cluster analysis is performed by combining the spatial location of anomalies with the consistency of causes to identify isolated point problems or potential regional events. Finally, the judgment threshold is dynamically adjusted using the data distribution of target viewing points, enabling the entire system to have continuous learning and optimization capabilities, forming a complete closed loop. This significantly improves the scientific, forward-looking, and refined level of maintenance management, effectively solving the problems of low accuracy in anomaly identification and delayed risk warning caused by over-reliance on static data.

[0034] Please see Figure 2The diagram illustrates the logic for determining suspicious viewing points in this embodiment. In this embodiment, the process of determining a landscape index based on the green visibility rate, color vitality value, and image transparency, and determining several suspicious viewing points based on the comparison between the landscape index and a preset index threshold, includes: performing maximum-minimum normalization on the green visibility rate to obtain a green visibility normalized value; performing maximum-minimum normalization on the color vitality value to obtain a color normalized value; performing maximum-minimum normalization on the image transparency to obtain a transparency normalized value; performing a weighted summation of the green visibility normalized value, the color normalized value, the transparency normalized value, a preset green visibility rate weight, a preset color weight, and a preset transparency weight to obtain a landscape index; and determining the viewing point to be tested as a suspicious viewing point when the landscape index is less than the preset index threshold, thereby identifying several suspicious viewing points.

[0035] The preset green view rate weight is an importance ratio coefficient assigned to the green view rate when calculating the comprehensive landscape index. It depends on the core management objectives and design style of the garden project and is usually between 0.3 and 0.5. In this embodiment, it is set to 0.5, which can emphasize the ecological and visual foundation of green coverage, conform to the core principle of garden greening that is based on green, and can prioritize the screening of the fundamental problem of insufficient green space.

[0036] The preset color weight is an importance ratio coefficient assigned to the color vitality value when calculating the comprehensive landscape index. It depends on the seasonal viewing needs and artistic design requirements of the garden, and is usually between 0.2 and 0.4. In this embodiment, it is set to 0.3, which highlights the ecological foundation while reasonably incorporating the contribution of color aesthetics to landscape quality, avoiding an overly simplistic evaluation system, and can identify the decline in aesthetics caused by monotonous or unbalanced colors.

[0037] The preset transparency weight is an importance ratio coefficient assigned to the transparency of the image when calculating the comprehensive landscape index. It depends on the spatial layout and functional zoning of the garden and is usually between 0.2 and 0.4. In this embodiment, it is set to 0.2 to ensure that the sense of spatial closure is adequately monitored while avoiding its excessive interference with the evaluation of the vegetation itself, so that the initial screening is more focused on the physical attributes of the vegetation.

[0038] By constructing a comprehensive evaluation model integrating ecology, aesthetics, and spatial perception, an objective diagnosis of landscape aesthetics is achieved. Green view rate directly reflects the ecological foundation and visual green volume of vegetation cover; color vitality value characterizes the color richness and visual appeal of the landscape; and image transparency relates to the sense of openness and visual comfort of the space. These three factors define the visual health status of the landscape from three dimensions: green volume, color, and space. Normalization eliminates dimensional differences in the original data, and pre-set weights are introduced for weighted summation, enabling the correlation and fusion of multi-dimensional heterogeneous data. The resulting landscape index becomes a unified and comparable benchmark. By comparing it with preset index thresholds, suspicious viewing points that fail to meet the comprehensive dimensions can be accurately identified, laying a reliable data foundation and avoiding misjudgments caused by the one-sidedness of a single indicator.

[0039] Specifically, the process of determining the type tendency value based on the temporal variation characteristics of the image transparency and prominence of each of the suspected viewing points includes: calculating the standard deviation of all image transparency from the initial time to each time within a preset judgment period to obtain several transparency fluctuation values; calculating the standard deviation of all prominence from the initial time to each time within the preset judgment period to obtain several prominence fluctuation values; performing max-min normalization on all transparency fluctuation values ​​to obtain several transparency standard values; performing max-min normalization on all prominence fluctuation values ​​to obtain several prominence standard values; and calculating the Pearson correlation coefficient between all transparency standard values ​​and all prominence standard values ​​to obtain the type tendency value.

[0040] The preset judgment duration is the length of time used to calculate the fluctuations of two indicators: image clarity and visual focus prominence. It depends on the natural cycle of landscape changes (such as plant growth rate and seasonal changes) and the frequency of image acquisition, and is usually set between 2 weeks and 3 months. In this embodiment, it is set to 30 days, which can effectively cover a relatively complete short-term growth or maintenance cycle. This ensures that there are enough data points for reliable statistical analysis, while also sensitively reflecting recent dynamics and avoiding interference from ultra-long-term trends.

[0041] By analyzing the coordinated fluctuation patterns of landscape visual features over time, intelligent prediction of landscape types was achieved. The fluctuations in image transparency and visual focal prominence respectively characterize the temporal stability of spatial openness and visual guidance. In ideal visual control areas, due to strong constraints from human planning, their fluctuations should show a high positive correlation. However, in spatial creation areas dominated by natural changes, the correlation between the two fluctuations is weaker. By normalizing the fluctuation sequence and calculating the Pearson correlation coefficient, this step quantifies the coordinated relationship between the two fluctuations over time into a scalar value, namely the type tendency value. Essentially, this constructs a mathematical indicator to measure the intensity of visual control, enabling a deep logical connection between static space and visual indicators at the dynamic data level. This allows for the automatic identification of landscape types that conform to management intentions from the data, providing an accurate classification basis for subsequent targeted anomaly diagnosis.

[0042] Please see Figure 3 As shown, this is the logic diagram for determining the landscape type in this embodiment. In this embodiment, the process of determining several viewpoints for line-of-sight control and several viewpoints for spatial creation based on the comparison results of the type tendency value and the preset tendency threshold includes: When the type tendency value is greater than or equal to the preset tendency threshold, the suspicious viewing point is determined to be the line-of-sight control type viewing point; when the type tendency value is less than the preset tendency threshold, the suspicious viewing point is determined to be the space creation type viewing point.

[0043] By intelligently mapping continuous data reflecting the intensity of visual control to discrete landscape categories with clear management semantics, a higher type tendency value indicates a highly coordinated fluctuation in transparency and prominence, revealing that landscape visual elements are in a controlled and interconnected state of change. This aligns with the professional characteristics of visual control areas, which require strict management of visual corridors and focal sequences. Conversely, a lower type tendency value indicates that the fluctuations of the two are relatively independent, reflecting a more natural and random landscape change. This aligns with the management goal of space creation areas, which pursue a diverse and flexible visual experience. The preset tendency threshold acts as a classification decision boundary, enabling the system to automatically and consistently transform data features into type labels that can directly guide differentiated maintenance decisions. Thus, after identifying a problem, it can accurately pinpoint the type of problem and determine the appropriate strategy for handling it, achieving a deep connection and closed loop between data flow and maintenance business flow.

[0044] Please see Figure 4As shown, this is the logic diagram for determining the cause of anomalies and obtaining abnormal viewing points in this embodiment. In this embodiment, the process of determining the cause of anomalies and obtaining several abnormal viewing points based on the occupancy rate and the visibility-based line-of-sight occupancy index includes: calculating the rate of change of the occupancy rate within a preset time period to obtain the occupancy rate change rate; calculating the rate of change of the visibility within the preset time period to obtain the visibility change rate; calculating the negative of the ratio of the visibility change rate to the occupancy rate change rate to obtain the line-of-sight occupancy index; when the line-of-sight occupancy index is greater than a preset occupancy threshold, determining that the cause of anomalies in the line-of-sight control type viewing point is plant occupancy, thereby determining the line-of-sight control type viewing point as the abnormal viewing point and obtaining several abnormal viewing points.

[0045] The preset duration is the length of time used to diagnose visual obstruction problems. It depends on the typical cycle in which plant growth (especially foliage) is sufficient to have an observable impact on the visual corridor, usually between 15 and 60 days. In this embodiment, it is set to 30 days, which can sensitively capture the significant changes brought about by the rapid growth period of foliage (such as in spring) without being disturbed by daily fluctuations due to the short time frame, providing a stable and practically meaningful observational basis for calculating reliable rates of change.

[0046] The preset occlusion threshold is a critical value used to determine whether the visibility occlusion index constitutes an abnormal plant invasion. It depends on the management's tolerance for significant vision deterioration and the quantitative standard for the relationship between visibility and invasion rate changes, and is typically set between 0.5 and 2.0. In this embodiment, it is set to 1.0, meaning that the relative rate of visibility decline must be at least equivalent to the relative rate of invasion rate increase. This effectively excludes misjudgments caused by other reasons and ensures that the diagnosed plant invasion abnormality has a clear necessity for maintenance intervention.

[0047] Based on the principle that a significant negative correlation should exist between the decrease in visibility and the increase in vegetation encroachment rate over time within the visual control area, the system calculates the rate of change of both and obtains the negative value of the ratio. This physical relationship is creatively transformed into a visual obstruction index: when plant growth does indeed constitute obstruction, the index will show a large positive value because the rate of decrease in visibility is negative and the rate of increase in encroachment rate is positive. Conversely, if the change in visibility is caused by other factors, the index will not rise abnormally. This establishes a strong logical connection between the trend of change in the data layer and the abnormal cause (plant encroachment) in the physical layer. By setting a preset obstruction threshold as an empirical critical point, the system can automatically filter out those points where the visibility deteriorates significantly due to the dynamic growth of vegetation. This allows the system to accurately locate general problems in visual control as abnormal viewing points that require intervention, achieving an intelligent leap from type judgment to cause diagnosis.

[0048] Specifically, the process of determining the cause of anomalies based on the spatial quality index calculated using the color vitality value and the morphological regularity, thereby obtaining several abnormal viewing points, includes: performing maximum-minimum normalization on the color vitality value to obtain a color standard value; performing maximum-minimum normalization on the morphological regularity to obtain a regularity standard value; constructing two-dimensional data points using the color standard value as the abscissa and the regularity standard value as the ordinate; calculating the Euclidean distance between the two-dimensional data points and a preset health benchmark point to obtain a health distance; and determining that the cause of anomalies at the spatially constructed viewing point is plant disqualification when the health distance is greater than a preset health distance threshold, thereby identifying the spatially constructed viewing point as the abnormal viewing point and obtaining several abnormal viewing points.

[0049] The preset health benchmark point is a coordinate point representing the ideal healthy landscape state of a garden within a two-dimensional aesthetic state space composed of color standard values ​​and regularity standard values. It depends on the specific garden's design goals, plant configuration standards, and maximum maintenance level, and its horizontal and vertical coordinates are typically set between 0.7 and 1.0. In this embodiment, it is set to (0.85, 0.90), setting high standards for both color and form. This aligns with the professional understanding that high-quality landscapes typically require more refined styling and maintenance, establishing a clear ideal benchmark for determining plant disqualification.

[0050] The preset health distance threshold is the critical distance for determining whether the current observation point's deviation from the ideal health benchmark constitutes a disqualification. It depends on the management's tolerance for aesthetic decline in the plant landscape and the quantitative definition of significant disqualification, and is typically set between 0.15 and 0.35. In this embodiment, it is set to 0.25, which avoids being too small to be overly sensitive to normal fluctuations and generate numerous false alarms, while also avoiding being too large to be overly insensitive and miss aesthetic decline issues that have clearly deviated from the healthy state.

[0051] By transforming subjective judgments of plant landscape disqualification into objective spatial distance measurements, color vitality and morphological regularity define the two core aesthetic dimensions of plant visual health: color and form. Together, they form a two-dimensional evaluation plane. After normalization to eliminate dimensions, the state of each viewing point is mapped to a data point on the plane. The preset health benchmark represents the ideal state of excellent color and form in the context of the garden. Calculating the Euclidean distance between the two essentially measures the comprehensive deviation between the current state and the ideal state, thus mathematically linking the two originally separate indicators. The geometric distance intuitively reflects the degree of decline in comprehensive aesthetic quality. When the distance exceeds the preset threshold, it indicates an unacceptable deviation in both color and form. The system can then accurately diagnose the plant as disqualified, thereby transforming the vague unattractive problem in the spatial creation area into a clear, measurable, and intervention-required specific abnormal cause.

[0052] Specifically, the process of determining several target viewing points based on the spatial location of each of the abnormal viewing points and the cause of the abnormality includes: calculating the Euclidean distance between the position coordinates of any two abnormal viewing points to obtain the abnormal distance, and marking the two abnormal viewing points as the first abnormal viewing point and the second abnormal viewing point, respectively; when the abnormal distance is greater than or equal to a preset abnormal distance threshold, determining the first abnormal viewing point and the second abnormal viewing point as the target viewing points to determine several target viewing points; when the abnormal distance is less than the preset abnormal distance threshold, determining several target viewing points based on the cause of the abnormality of the first abnormal viewing point and the cause of the abnormality of the second abnormal viewing point.

[0053] The preset anomaly distance threshold is a critical distance value used to determine whether two anomalous viewing points are spatially adjacent and thus potentially related. It depends on the spatial scale of the garden, the visual relevance of the landscape, and the typical diffusion range of common pests, diseases, or maintenance problems, and is typically set between 20 and 50 meters. In this embodiment, it is set to 30 meters, which effectively captures nearby anomalies that may be caused by the same reason, while avoiding misjudging distant, highly independent anomalies as related events, thus ensuring the accuracy and practicality of spatial clustering analysis.

[0054] By introducing spatial correlation analysis, discrete abnormal alarm points are upgraded to problem areas with management significance. By calculating the Euclidean distance between abnormal points and comparing it with a preset threshold, the spatial density and diffusion probability of abnormal events are first quantified at the data level. When the distance between two points is less than the threshold, it means that they are highly correlated in physical space. At this time, further comparison of the causes of the anomalies becomes the key basis for decision-making, enabling the system to intelligently distinguish between sporadic cases and potential regional risks. This optimizes maintenance decisions from responding to single-point anomalies to proactively identifying and intervening in problem patterns, realizing data-driven, precise and efficient resource scheduling and early warning upgrades.

[0055] Specifically, the process of determining several target viewing points based on the causes of the first and second abnormal viewing points includes: when the causes of the first and second abnormal viewing points are different, determining the first and second abnormal viewing points as target viewing points to determine several target viewing points; when the causes of the first and second abnormal viewing points are the same, determining that an abnormal regional event has occurred; based on the abnormal regional event, calculating the geometric center of the position coordinates of the first and second abnormal viewing points to obtain the regional center; taking the regional center as the center and the area of ​​a circle obtained according to a preset abnormal radius as the abnormal region, determining all the abnormal viewing points within the abnormal region as target viewing points to determine several target viewing points; extracting the landscape index of all the viewing points to be measured within the abnormal region, calculating the rate of change of each landscape index within a preset observation period to obtain the index change rate; and determining the viewing point to be measured as the target viewing point when the index change rate is greater than a preset landscape change threshold to determine several target viewing points.

[0056] The preset anomaly radius is the radius of a circle used to delineate the abnormal area with the center of the area as the center when an abnormal event is determined to occur. It depends on the typical physical diffusion or influence range of the determined anomaly cause, and is usually between 10 meters and 30 meters. In this embodiment, it is set to 15 meters, which can effectively cover a reasonable risk buffer outside the two anomaly points, and is used to capture potential risk points that may have been affected but have not yet reached the anomaly standard, thereby achieving early intervention.

[0057] The preset landscape change threshold is a critical value used to determine whether the rate of change of the landscape index at a test viewing point within an abnormal area has reached the standard of significant deterioration. It depends on the natural fluctuation range of the landscape index under normal conditions and is usually set between 0.10 and 0.25. In this embodiment, it is set to 0.15, which can effectively capture significant downward trends that exceed normal fluctuations, promptly include points showing signs of deterioration in the early warning, and at the same time reduce unnecessary false alarms.

[0058] By integrating three types of data—spatial location, cause of anomalies, and temporal changes—a dynamic risk assessment model was constructed. When adjacent anomalies share the same cause, they are identified as events originating from the same region. The potential impact range is delineated using its geometric center and a preset radius, which aligns with the objective law of the continuous spatial spread of problems such as pests, diseases, and improper maintenance. Furthermore, the system not only monitors the anomalies already discovered within the range but also all points to be monitored. By calculating the rate of change of their landscape indices and comparing it with thresholds, it essentially checks whether healthy points within the monitoring area show an early trend towards anomalies. This creates a deep correlation between spatial clustering information, problem attribute information, and temporal early warning signals, enabling the system to move beyond passively recording existing problems and proactively capture the vanguard of potential spread. This shifts the determination of target viewing points from dealing with existing anomalies to intercepting developing risks, achieving optimal allocation and forward-looking deployment of maintenance resources.

[0059] Specifically, the process of adjusting the preset index threshold based on the numerical distribution characteristics of the landscape index at each target viewing point includes: calculating the average value of all the landscape indices to obtain the average landscape index; calculating the relative deviation between the average landscape index and the preset index threshold to obtain the landscape deviation; and when the landscape deviation is greater than the preset landscape deviation threshold, reducing the preset index threshold based on the relative deviation between the landscape deviation and the preset landscape deviation threshold and a preset adjustment coefficient, where Q'=Q×[1-a×|F-F'| / F'], Q' is the adjusted preset index threshold, Q is the preset index threshold, a is the preset adjustment coefficient, F is the landscape deviation, and F' is the preset landscape deviation threshold.

[0060] The preset landscape deviation threshold is a critical value used to determine whether the landscape deviation is significant enough to trigger threshold adjustment. It depends on the system's definition of a significant discrepancy between the standard and reality, as well as the requirements for threshold stability, and is usually set between 0.1 and 0.2. In this embodiment, it is set to 0.15, which can effectively avoid frequent and unnecessary threshold adjustments due to normal fluctuations or a few extreme values ​​in a single monitoring data session, thus ensuring the stability of the system's core evaluation criteria.

[0061] The preset adjustment coefficient is a proportional coefficient used to control the magnitude of the index threshold adjustment. It depends on the balance between the desired system adaptive speed and stability, and is usually between 0.1 and 0.3. In this embodiment, it is set to 0.2, so that the threshold can smoothly and gradually adapt to changes in the overall landscape level, effectively preventing the impact on the system's judgment logic caused by a single drastic adjustment, and achieving an adaptive effect of slow learning and steady optimization.

[0062] By calculating the average landscape index of the selected target viewing points, a data benchmark reflecting the typical level of the current substandard group is obtained. The relative deviation of this average value from the current standard is calculated, which essentially quantifies the gap between common problems and management requirements. When this deviation exceeds the sensitivity threshold, a threshold adjustment based on a preset coefficient is triggered. This is because if the index of a large number of problem points is concentrated at a level significantly lower than the current threshold, it may indicate that the existing standard has deviated from the generally achievable state of the landscape, or that the overall maintenance level has undergone a phased change. By automatically and gradually lowering the threshold, the system can avoid continuously generating a large number of invalid alarms due to rigid standards, and keep the monitoring focus in sync with the achievable and generally healthy state of the garden. This makes the anomaly judgment more practically instructive and provides a self-calibrating reference benchmark for the dynamic allocation of maintenance resources.

[0063] Specifically, the process of adjusting the preset tendency threshold based on the numerical distribution characteristics of the type tendency values ​​of each target viewing point includes: clustering all the type tendency values ​​to obtain a first candidate cluster and a second candidate cluster; calculating the median of all the type tendency values ​​within the first candidate cluster to obtain a first typical tendency value; calculating the median of all the type tendency values ​​within the second candidate cluster to obtain a second typical tendency value; calculating a classification reference value based on the first typical tendency value and the second typical tendency value; calculating the relative deviation between the classification reference value and the preset tendency threshold to obtain a tendency deviation; and adjusting the preset tendency threshold based on the relative deviation between the tendency deviation and the preset tendency deviation threshold, and a preset adjustment coefficient, when the absolute value of the tendency deviation is greater than the preset tendency deviation threshold. , The adjusted preset tendency threshold, To preset the tendency threshold, The preset adjustment coefficient, This is a tendency bias. This is a preset tendency deviation threshold.

[0064] Formula for calculating classification reference values: Where R is the classification reference value, N1 is the number of samples in the first candidate cluster, N2 is the number of samples in the second candidate cluster, M1 is the first typical tendency value, and M2 is the second typical tendency value.

[0065] The preset bias bias threshold is a critical value used to determine whether the bias bias is significant enough to trigger adjustment. It depends on the system's requirements for the stability of the classification boundary and its tolerance for data noise, and is usually set between 0.05 and 0.15. In this embodiment, it is set to 0.10, which means that the system only considers the classification criteria to need optimization when there is a clear and continuous deviation between the data-driven theoretical boundary and the current application boundary.

[0066] The preset adjustment coefficient is a proportional coefficient used to control the magnitude of the tendency threshold adjustment. It depends on the desired rate of evolution of the classification rules and is usually set between 0.1 and 0.3. In this embodiment, it is set to 0.2, which allows the classification rules to smoothly and slowly track and adapt to the long-term, continuous drift of landscape type characteristics caused by plant growth, seasonal changes, or changes in maintenance focus. This avoids abrupt adjustments and maintains the consistency of the entire monitoring system's discrimination logic.

[0067] By performing unsupervised clustering on the type tendency values ​​of identified outliers, the natural distribution centers of two types of problems—sight control and space creation—were discovered in the current data; these are identified as the first and second typical tendency values. The weighted average of these two center values ​​(classification reference value) was used as the theoretically optimal classification boundary suggested by the current data. By calculating the relative deviation between this theoretical boundary and the current application boundary (preset tendency threshold), the degree of mismatch between the real-world problem pattern and the existing classification rules was quantified. When the absolute value of the deviation exceeded the sensitivity threshold, the threshold was fine-tuned to ensure that the classification rules continuously approached the optimal solution revealed by the data. This established a strong correlation between the abstract landscape type determination and the actual distribution of specific, dynamically changing outliers, achieving closed-loop self-optimization of the classifier's performance in application. This ensured that the system could continuously adapt to the long-term impacts of plant growth, seasonal changes, and shifts in maintenance priorities.

[0068] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A landscape greening monitoring method based on image recognition, characterized in that, include: Real-time acquisition of greening images of various viewing points in the landscape garden, and extraction of greening rate, color vitality value, image transparency, morphological regularity, vegetation encroachment rate of the visual corridor, prominence of visual focus and visibility from the greening images. The landscape index is determined based on the green visibility rate, the color vitality value, and the image transparency, and several suspicious viewing points are identified based on the comparison results between the landscape index and the preset index threshold. Based on the temporal variation characteristics of the image transparency and prominence of each of the suspected viewing points, a type tendency value is determined, and based on the comparison results of the type tendency value and the preset tendency threshold, a number of view control type viewing points and a number of space creation type viewing points are determined. Based on the aforementioned viewpoints with controlled line of sight, the cause of the anomaly is determined by the line of sight obstruction index calculated according to the occupancy rate and the visibility, thereby obtaining several abnormal viewpoints. Based on the aforementioned spatial creation viewing points, the causes of anomalies are determined by the spatial quality index calculated according to the color vitality value and the morphological regularity, thereby obtaining several abnormal viewing points. Based on the spatial location of each abnormal viewing point and the cause of the abnormality, several target viewing points are determined; The preset index threshold is adjusted according to the numerical distribution characteristics of the landscape index of each target viewing point, and the preset tendency threshold is adjusted according to the numerical distribution characteristics of the type tendency value of each target viewing point; Greening intervention early warning is issued for all target viewpoints obtained after adjusting the preset index threshold and the preset tendency threshold.

2. The landscape and greening monitoring method based on image recognition according to claim 1, characterized in that, The process of determining a landscape index based on the green visibility rate, the color vitality value, and the image clarity, and then identifying several suspicious viewing points based on a comparison between the landscape index and a preset index threshold, includes: The landscape index is calculated based on the green visibility rate, the color vitality value, and the image transparency. When the landscape index is less than the preset index threshold, the viewing point to be tested is determined as a suspicious viewing point, thereby identifying several suspicious viewing points.

3. The landscape greening monitoring method based on image recognition according to claim 2, characterized in that, The process of determining the type tendency value based on the temporal variation characteristics of the image clarity and prominence of each of the aforementioned suspicious viewing points includes: Calculate several transparency fluctuation values ​​based on the image transparency within a preset judgment time period; Calculate several prominence fluctuation values ​​based on the prominence within the preset determination time period; The type tendency value is calculated based on all the transparency fluctuation values ​​and all the prominence fluctuation values.

4. The landscape and greening monitoring method based on image recognition according to claim 3, characterized in that, The process of determining several viewpoints for line-of-sight control and several viewpoints for spatial creation based on the comparison results of type tendency values ​​and preset tendency thresholds includes: When the type tendency value is greater than or equal to the preset tendency threshold, the suspicious viewing point is determined to be the line-of-sight control type viewing point; when the type tendency value is less than the preset tendency threshold, the suspicious viewing point is determined to be the space creation type viewing point.

5. The landscape and greening monitoring method based on image recognition according to claim 4, characterized in that, The process of determining the cause of the anomaly based on the occupancy rate and the visibility-calculated visual obstruction index, thereby obtaining several abnormal viewing points, includes: Calculate the rate of change of occupancy rate based on the occupancy rate within a preset time period; Calculate the visibility change rate based on the visibility within the preset time period; Calculate the line obstruction index based on the occupancy rate change rate and the visibility change rate; Based on the comparison between the line-of-sight obstruction index and the preset obstruction threshold, the abnormality of the line-of-sight control viewing point is determined to be plant encroachment, thereby identifying the line-of-sight control viewing point as the abnormal viewing point and obtaining several abnormal viewing points.

6. The landscape and greening monitoring method based on image recognition according to claim 5, characterized in that, The process of determining the cause of anomalies and obtaining several abnormal viewing points based on the spatial quality index calculated from the color vitality value and the morphological regularity includes: Two-dimensional data points are constructed based on the color vibrancy value and the shape regularity. Calculate the health distance based on the two-dimensional data points and the preset health benchmark points; Based on the comparison between the healthy distance and the preset healthy distance threshold, the abnormality of the spatially designed viewing point is determined to be plant disqualification, thereby identifying the spatially designed viewing point as the abnormal viewing point and obtaining several abnormal viewing points.

7. The landscape and greening monitoring method based on image recognition according to claim 6, characterized in that, The process of determining several target viewing points based on the spatial location of each of the aforementioned abnormal viewing points and the cause of the abnormality includes: Calculate the abnormal distance based on the position coordinates of any two abnormal viewing points, and mark the two abnormal viewing points as the first abnormal viewing point and the second abnormal viewing point, respectively. When the abnormal distance is greater than or equal to a preset abnormal distance threshold, the first abnormal viewing point and the second abnormal viewing point are determined as the target viewing point, so as to determine a number of target viewing points; When the abnormal distance is less than the preset abnormal distance threshold, several target viewing points are determined based on the abnormal cause of the first abnormal viewing point and the abnormal cause of the second abnormal viewing point.

8. The landscape and greening monitoring method based on image recognition according to claim 7, characterized in that, The process of determining several target viewing points based on the causes of the anomalies of the first and second abnormal viewing points includes: Based on the first comparison result of the abnormal cause of the first abnormal viewing point and the abnormal cause of the second abnormal viewing point, the first abnormal viewing point and the second abnormal viewing point are determined as the target viewing points, thereby identifying a number of target viewing points; Based on the second comparison result of the abnormal cause of the first abnormal viewing point and the abnormal cause of the second abnormal viewing point, an abnormal event in the area is determined; Based on the regional anomaly events, the regional center is determined according to the first and second anomaly viewing points. An abnormal region is determined based on the region center and a preset abnormal radius. All abnormal viewing points within the abnormal region are identified as target viewing points to determine a number of target viewing points. The landscape index of all the viewing points to be tested within the abnormal region is extracted. The rate of change of each landscape index is calculated. When the rate of change of the index is greater than a preset landscape change threshold, the viewing point to be tested is identified as the target viewing point to determine a number of target viewing points.

9. The landscape and greening monitoring method based on image recognition according to claim 8, characterized in that, The process of adjusting the preset index threshold based on the numerical distribution characteristics of the landscape index at each target viewing point includes: Calculate the mean of the landscape indices based on all the landscape indices mentioned; The landscape deviation is calculated based on the mean of the landscape index and the preset index threshold. The preset index threshold is adjusted based on the comparison between the landscape deviation and the preset landscape deviation threshold.

10. The landscape and greening monitoring method based on image recognition according to claim 9, characterized in that, The process of adjusting the preset tendency threshold based on the numerical distribution characteristics of the type tendency values ​​of each target viewing point includes: Clustering is performed on all the aforementioned type tendency values ​​to obtain a first candidate cluster and a second candidate cluster; Calculate the median of all the type tendency values ​​within the first candidate cluster to obtain the first typical tendency value; Calculate the median of all the type tendency values ​​within the second candidate cluster to obtain the second typical tendency value; A classification reference value is calculated based on the first typical tendency value and the second typical tendency value. A tendency deviation is calculated based on the classification reference value and a preset tendency threshold. The preset tendency threshold is adjusted based on the comparison between the tendency deviation and the preset tendency deviation threshold.

Images