Sample classification and dynamic updating method for environmental protection remote sensing images of power transmission and transformation project

Abstract

The application relates to the technical field of intelligent management of a remote sensing image sample library of a power transmission and transformation project, and discloses a method for classifying and dynamically updating remote sensing images of a power transmission and transformation project. According to the method, a vitality file is established for each sample, and classification is performed according to three-dimensional rules of terrain, water environment protection targets and project violation types. Based on this, an ecological pressure and ecological vitality distribution map is calculated, and a dynamic mechanism for competition and cooperation of newly-added samples in the library is guided. Meanwhile, periodic vitality decay and elimination strategies based on the distribution map are adopted, so that the power transmission and transformation project sample library of the water environment protection is automatically updated and optimized. The application solves the technical problems that the existing static power transmission and transformation project sample library of the water environment protection cannot adapt to project changes, the sample quality is uneven, and the sample library lacks self-optimization capability, so that the generalization and accuracy of a target recognition model are insufficient, and the quality of the power transmission and transformation project sample library of the water environment protection is continuously and autonomously improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent management technology for remote sensing image sample libraries, and in particular to a method for classifying and dynamically updating remote sensing image samples for environmental protection and water conservation in power transmission and transformation projects. Background Technology

[0002] In environmental protection and soil and water conservation work for power transmission and transformation projects, the automatic identification and monitoring of targets using remote sensing imagery has become an important technical means. The core foundation of this technology is a high-quality, representative environmental and soil conservation remote sensing imagery database for power transmission and transformation projects. In existing technologies, the construction and maintenance of such databases largely rely on manual experience, typically employing a static collection and one-time injection model, with subsequent updates mostly consisting of simple incremental additions. This approach lacks a systematic and dynamic evaluation and management of sample quality, inter-sample relationships, and the distribution structure within the database.

[0003] Current methods suffer from the following fundamental flaws: Firstly, the static environmental and water conservation sample database for power transmission and transformation projects cannot dynamically adapt to changes in sample feature requirements under different terrains, construction stages, and the emergence of new violation types. This limits the generalization ability of models trained on this database. Secondly, the database lacks self-cleaning and evolutionary mechanisms, leading to the continuous accumulation of low-quality, redundant, or outdated samples, squeezing out high-quality samples, while scarce key samples are not effectively replenished. This results in the overall degradation of the database's quality, severely restricting the accuracy and reliability of subsequent intelligent recognition algorithms. Therefore, there is an urgent need for a management method for the environmental and water conservation sample database of power transmission and transformation projects that can simulate the dynamic balance of an ecosystem and achieve autonomous selection, intelligent replenishment, and structural optimization of samples.

[0004] Therefore, this invention proposes a method for classifying and dynamically updating remote sensing image samples for water conservation in power transmission and transformation projects. Summary of the Invention

[0005] This invention provides a method for classifying and dynamically updating remote sensing image samples of environmental protection and water conservation in power transmission and transformation projects. It overcomes the shortcomings of traditional static environmental protection and water conservation sample databases for power transmission and transformation projects, which suffer from sample quality degradation and insufficient adaptability due to the lack of dynamic optimization mechanisms. By simulating the dynamic balance of the ecosystem, it realizes the autonomous evolution and continuous optimization of the environmental protection and water conservation sample database for power transmission and transformation projects.

[0006] This invention provides a method for classifying and dynamically updating remote sensing image samples related to environmental protection and water conservation in power transmission and transformation projects, including: Construct a sample library of environmental protection and water conservation for power transmission and transformation projects, and create and maintain a sample lifespan profile for each sample in the sample library; The classification space coordinates of each sample are determined according to the preset three-dimensional classification rules, which include the terrain type dimension, the environmental protection and water conservation target dimension, and the engineering violation type dimension. Based on the classification spatial coordinates and sample vitality profiles of all samples, ecological pressure distribution maps and ecological vitality distribution maps of the environmental protection and water conservation sample bank for power transmission and transformation projects are calculated and generated. Receive new remote sensing image samples and generate sample vitality profiles and classification spatial coordinates for the new samples; Based on the classification spatial coordinates of the new samples, the first pressure value is extracted from the ecological pressure distribution map, and the first vitality value is extracted from the ecological vitality distribution map. Convert the first pressure value into a competition intensity coefficient and the first vitality value into a synergy intensity coefficient; Based on the competition intensity coefficient, the competitive entry process is executed: determine whether the new sample wins the competition, and if so, accept the new sample and adjust the sample lifespan profile of the competing samples in the database. Based on the collaboration strength coefficient, the collaboration gain process is executed: determine whether the new sample triggers collaboration, and if so, accept the new sample and improve the sample life file of the new sample and the collaborative samples in the library. Perform a vitality decay operation on the sample vitality profiles of all samples periodically; Based on the weakened sample viability profile, a sample elimination operation is performed to remove samples with weak viability, and the ecological pressure distribution map and ecological vitality distribution map are recalculated. Output the updated sample library for environmental protection and water conservation in power transmission and transformation projects.

[0007] Preferably, the sample vitality profile consists of three sub-items: vitality score, sample entropy value, and sample popularity. The vitality score is calculated based on the clarity of the sample image, the consistency of the annotation, and the typicality of the visual features. The sample entropy value is calculated based on the distance between the visual features of a sample and the cluster center of similar sample features. The greater the distance, the higher the sample entropy value. The sample popularity record shows the number of times the sample has been successfully retrieved or accessed in history.

[0008] Preferably, the calculation of the ecological pressure distribution map includes: The three-dimensional classification space is divided into multiple uniform grid units; For each grid cell, count the number of samples within the cell and calculate the cell sample density index based on the number of samples. For each grid cell, the average of the vitality scores of all samples within the cell is calculated, and the cell quality competition index is calculated based on this average. The unit sample density index and the unit quality competition index of each grid cell are weighted and summed to obtain the unit ecological pressure value of the corresponding grid cell. The set of ecological pressure values ​​for all grid cells constitutes an ecological pressure distribution map.

[0009] Preferably, the calculation of the ecological vitality distribution map includes: The three-dimensional classification space is divided into multiple uniform grid units; Set a threshold for the number of samples, and mark grid cells with a sample number lower than the threshold as sparse cells; Analyze the logical correlation between each sparse unit and its adjacent non-sparse units in terms of terrain type and engineering stage, and calculate the unit correlation strength based on the logical correlation. Based on the unit association strength, the unit ecological vitality value that can be obtained by introducing new samples into the corresponding sparse unit is calculated. The higher the unit association strength, the greater the unit ecological vitality value. The set of unit ecological vitality values ​​of all grid cells constitutes the ecological vitality distribution map, and the unit ecological vitality value of non-sparse cells is zero.

[0010] Preferably, the conversion of the first pressure value into a competition intensity coefficient is based on a first preset mapping relationship, which defines that the larger the first pressure value, the larger the competition intensity coefficient obtained.

[0011] The preferred competitive warehousing process includes: The competition radius is determined based on the competition intensity coefficient, and the competition radius is inversely proportional to the competition intensity coefficient. The competition region is defined as the spatial region within the competition radius centered on the classification spatial coordinates of the newly added samples. Within the competitive region, find the existing sample with the highest vitality score in the sample vitality profile and use it as the direct competing sample; Compare the viability scores of newly added samples with those of directly competing samples; If the viability score of the new sample is higher than that of the directly competing sample, the new sample is deemed to have won the competition and is accepted. The viability score of the directly competing sample is then reduced based on the competition intensity coefficient.

[0012] Preferred, the synergistic gain process includes: The coordination radius is determined based on the coordination strength coefficient, which is directly proportional to the coordination radius. Using the classification spatial coordinates of the newly added samples as the center, a spatial region within the collaborative radius is defined as the collaborative region; Within the collaborative region, existing samples that are complementary to the new samples in terms of environmental protection and water conservation target types or engineering violation types are identified and used as collaborative gain samples. If a collaborative gain sample is found, it is determined that the new sample triggers collaboration, the new sample is accepted, and the vitality score of the new sample and all collaborative gain samples is increased according to the collaboration strength coefficient.

[0013] Preferably, the vitality decay operation follows an exponential decay model, where the vitality score in the sample vitality profile decreases exponentially with time intervals. The decay rate factor is set according to the environmental sensitivity level associated with the engineering violation type labeled in the sample. The higher the environmental sensitivity level, the smaller the decay rate factor, and the slower the vitality score decreases.

[0014] Preferably, the sample elimination operation includes: Different survival thresholds are set for grid cells with different ecological pressure values ​​in the ecological pressure distribution map. The higher the ecological pressure value of a cell, the higher the corresponding survival threshold. Traverse each grid cell and compare the vitality score in the vitality profile of the samples in that cell with the cell survival threshold. Samples with a vitality score lower than the survival threshold of their respective grid cells are marked as samples to be eliminated. When eliminating samples, they are removed from the power transmission and transformation project environmental protection sample pool in order of increasing sample entropy value, with samples of low entropy value being eliminated first.

[0015] Preferred options also include: Obtain the current ecological pressure distribution map and ecological vitality distribution map of the environmental protection and water conservation sample database for power transmission and transformation projects; In the ecological pressure distribution map, areas where the unit ecological pressure value is consistently higher than the high-pressure warning threshold are marked as high-pressure bottleneck areas. In the ecological vitality distribution map, areas where the unit ecological vitality value is consistently higher than the high potential warning threshold are marked as high potential sparse areas. Analyze the sample vitality profiles of samples within the high-pressure bottleneck area, and statistically analyze the proportion of samples with low vitality scores and the proportion of samples with high entropy values. Analyze the engineering semantics associated with high-potential sparse areas to determine the expected terrain type, environmental protection target type, and engineering violation type of the samples to be supplemented; Generate an optimization report for the environmental protection and water conservation sample library of power transmission and transformation projects. The optimization report includes the location information of high-voltage bottleneck areas, the sample quality analysis results within the areas, the location information of high-potential sparse areas, and the expected characteristic description of samples to be supplemented. Based on the optimization report, optimization suggestions for the environmental protection and water conservation sample library of power transmission and transformation projects are output. The optimization suggestions include suggestions for cleaning up samples in high-voltage bottleneck areas and suggestions for targeted sample collection and labeling in high-potential sparse areas.

[0016] The beneficial effects of this invention compared to existing technologies are as follows: Existing technologies have fundamental defects in the construction and maintenance of environmental protection and water conservation sample databases for power transmission and transformation projects using remote sensing images. Traditional methods typically employ a static, one-time collection and annotation model, with subsequent updates merely involving simple incremental overlays. This results in the power transmission and transformation project environmental protection and water conservation sample database failing to dynamically adapt to changes in the engineering environment and violation types, and lacking an inherent quality control and structural optimization mechanism. Consequently, the sample database accumulates a large number of redundant, low-quality, or outdated samples, while crucial scarce samples remain unreplenished, ultimately leading to weak generalization ability and decreased recognition accuracy of the intelligent recognition model trained based on this sample database. This invention, by introducing a vitality archive mimicking the dynamic balance of an ecosystem, a competition-cooperative data entry mechanism, and a periodic decay and elimination strategy based on a pressure-vitality distribution map, enables the power transmission and transformation project environmental protection and water conservation sample database to autonomously undergo selection and elimination, intelligently fill gaps, and optimize its internal structure, thereby completely overcoming the core problems of quality degradation and poor adaptability of static power transmission and transformation project environmental protection and water conservation sample databases.

[0017] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in this application.

[0018] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0019] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings: Figure 1 This is an overview diagram of the core method of the remote sensing image sample classification and dynamic updating method for water conservation in power transmission and transformation projects in this embodiment of the invention; Figure 2 This is a schematic diagram of the sample entry decision mechanism in an embodiment of the present invention; Figure 3 This is a flowchart illustrating the maintenance and self-optimization cycle of the environmental protection and water conservation sample library for power transmission and transformation projects in this embodiment of the invention. Detailed Implementation

[0020] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0021] like Figure 1 , Figure 2 , Figure 3As shown, this invention provides a method for classifying and dynamically updating remote sensing image samples for environmental protection and water conservation in power transmission and transformation projects, including: Construct a sample library of environmental protection and water conservation for power transmission and transformation projects, and create and maintain a sample lifespan profile for each sample in the sample library; The classification space coordinates of each sample are determined according to the preset three-dimensional classification rules, which include the terrain type dimension, the environmental protection and water conservation target dimension, and the engineering violation type dimension. Based on the classification spatial coordinates and sample vitality profiles of all samples, ecological pressure distribution maps and ecological vitality distribution maps of the environmental protection and water conservation sample bank for power transmission and transformation projects are calculated and generated. Receive new remote sensing image samples and generate sample vitality profiles and classification spatial coordinates for the new samples; Based on the classification spatial coordinates of the new samples, the first pressure value is extracted from the ecological pressure distribution map, and the first vitality value is extracted from the ecological vitality distribution map. Convert the first pressure value into a competition intensity coefficient and the first vitality value into a synergy intensity coefficient; Based on the competition intensity coefficient, the competitive entry process is executed: determine whether the new sample wins the competition, and if so, accept the new sample and adjust the sample lifespan profile of the competing samples in the database. Based on the collaboration strength coefficient, the collaboration gain process is executed: determine whether the new sample triggers collaboration, and if so, accept the new sample and improve the sample life file of the new sample and the collaborative samples in the library. Perform a vitality decay operation on the sample vitality profiles of all samples periodically; Based on the weakened sample viability profile, a sample elimination operation is performed to remove samples with weak viability, and the ecological pressure distribution map and ecological vitality distribution map are recalculated. Output the updated sample library for environmental protection and water conservation in power transmission and transformation projects.

[0022] In this embodiment, the sample vitality profile is a core data unit used to dynamically quantify and manage the value of each sample. It is not fixed, but is composed of a dynamic score reflecting the quality of the sample itself, an entropy value measuring the uniqueness of its characteristics, and a popularity that records the frequency of its use. It serves as a direct basis for the competition and elimination of samples in the database.

[0023] In this embodiment, the preset three-dimensional classification rules are the basis for constructing a structured and computable spatial index system. It maps the three key attributes of each sample—terrain environment, environmental protection target category, and engineering violation type—to a three-dimensional coordinate space, thereby providing a logical framework for subsequent analysis and management based on spatial relationships.

[0024] In this embodiment, the ecological pressure distribution map and the ecological vitality distribution map are global diagnostic maps generated based on sample spatial coordinates and vitality archives. The former is used to quantify the saturation level and internal competition intensity of samples in different classification regions, while the latter is used to actively identify high-potential regions with sparse samples but strong semantic correlations, jointly guiding the structural optimization of the environmental and water conservation sample library for power transmission and transformation projects.

[0025] In this embodiment, the competitive inclusion process and the collaborative gain process constitute the differentiated admission mechanism for new samples. The former simulates survival of the fittest, where in high-pressure areas where samples are saturated, new samples must directly compete with existing high-quality samples, and only the winner can be included in the database. The latter simulates symbiotic gain, where in high-potential areas that need to be supplemented, if a new sample can find existing samples with complementary features, they can collaboratively enhance each other's value and be strengthened together.

[0026] In this embodiment, the vitality decay operation simulates the natural depreciation process of sample value. The vitality score of the sample decays exponentially over time, and a slower decay rate is set for samples marked as high environmental sensitivity violations, thereby ensuring the timeliness of the environmental protection sample library for power transmission and transformation projects and the durability of key samples.

[0027] In this embodiment, the sample elimination operation is a dynamic cleanup mechanism based on the decayed vitality profile and ecological pressure distribution map. It sets differentiated survival thresholds in areas with different levels of crowding, and prioritizes the elimination of samples with low vitality and mediocre characteristics, thereby continuously releasing resources and optimizing the sample structure in the library.

[0028] To achieve a multi-dimensional quantitative assessment of sample quality and scarcity, and thus provide a precise basis for dynamic sample management, a sample vitality profile is proposed, consisting of three sub-items: vitality score, sample entropy value, and sample popularity. The vitality score is calculated based on the clarity of the sample image, the consistency of the annotation, and the typicality of the visual features. The sample entropy value is calculated based on the distance between the visual features of a sample and the cluster center of similar sample features. The greater the distance, the higher the sample entropy value. The sample popularity record shows the number of times the sample has been successfully retrieved or accessed in history.

[0029] In this embodiment, the vitality score is a comprehensive quality index calculated by integrating the quality attributes of the image itself. Image sharpness reflects the usability of the original data, annotation consistency measures the accuracy and reliability of manual annotation, and visual feature typicality assesses whether the visual pattern presented by the sample is generally representative of the category. For example, a sample with a blurry image, annotation bounding boxes that deviate significantly from the real object, and displaying atypical scenes in terms of lighting, angle, etc., will have a lower vitality score.

[0030] In this embodiment, the sample entropy value is a quantitative indicator that measures the uniqueness of sample features. It is obtained by calculating the distance between the visual feature vector of a specific sample and the average center (i.e., cluster center) of the features of all samples in its category. The larger the distance, the more the sample's features deviate from the mainstream pattern of its category, and the stronger its uniqueness or anomaly, thus resulting in a higher entropy value. For example, in the category of soil erosion in Taji, a sample exhibiting a rare erosion pattern will have a higher entropy value than a sample exhibiting a common erosion pattern.

[0031] In this embodiment, sample popularity is a dynamic indicator that records the practical value of a sample. It counts the number of times the sample has been successfully retrieved or invoked by model training, validation, or query tasks within a historical period. Sample popularity directly reflects the frequency of the sample's contribution to supporting the actual task. A frequently used sample indicates that the information it contains is of high value to the current task model, and its popularity will accumulate accordingly. This provides a direct basis for evaluating sample value based on usage frequency.

[0032] To quantitatively assess the crowding and quality competition intensity of samples within the environmental and water conservation sample bank for power transmission and transformation projects, and thus accurately identify redundant or inefficient areas, a method for calculating ecological pressure distribution maps is proposed, including: The three-dimensional classification space is divided into multiple uniform grid units; For each grid cell, count the number of samples within the cell and calculate the cell sample density index based on the number of samples. For each grid cell, the average of the vitality scores of all samples within the cell is calculated, and the cell quality competition index is calculated based on this average. The unit sample density index and the unit quality competition index of each grid cell are weighted and summed to obtain the unit ecological pressure value of the corresponding grid cell. The set of ecological pressure values ​​for all grid cells constitutes an ecological pressure distribution map.

[0033] In this embodiment, dividing the three-dimensional classification space into multiple uniform grid units is a process of discretizing and gridding the continuous classification coordinate space. This step is similar to drawing a latitude and longitude grid on a map, with each grid representing a small area containing a specific combination of terrain, environmental protection targets, and violation types. This operation forms the basis for subsequent spatial statistical analysis, enabling the system to partition, measure, and compare the classification space.

[0034] In this embodiment, the unit sample density index and the unit quality competition index are the two core dimensions constituting the unit ecological pressure value. The unit sample density index directly reflects the saturation level of the number of samples within the grid unit; the more samples there are, the higher the density index. The unit quality competition index reflects the average quality level of the samples within the grid unit; the higher the average viability score, the stronger the overall quality of existing samples in the area, and the greater the competitive pressure faced by new or low-quality samples. By weighted summing of these two indices, the unit ecological pressure value can comprehensively characterize the intensity of competition and access difficulty in an area caused by both an excessive number and high quality of samples.

[0035] In this embodiment, the ecological pressure distribution map is a spatial set of ecological pressure values ​​for all grid cells, visualized or quantified. This map intuitively or quantitatively shows which regions in the entire three-dimensional classification space have redundant and high-quality samples (high-pressure areas), and which regions have sparse or average-quality samples (low-pressure areas). For example, if a grid cell representing plain-construction road-dust pollution is shown as a high-pressure area, it indicates that the environmental protection sample bank for this category of power transmission and transformation projects is already highly saturated and generally of high quality. The system should suppress the blind addition of low-quality new samples and may trigger a sample elimination mechanism in this area.

[0036] To proactively identify potentially high-value areas with sparse but highly correlated samples, thereby guiding the intelligent and targeted replenishment of the environmental and water conservation sample bank for power transmission and transformation projects, a computational ecological vitality distribution map is proposed, including: The three-dimensional classification space is divided into multiple uniform grid units; Set a threshold for the number of samples, and mark grid cells with a sample number lower than the threshold as sparse cells; Analyze the logical correlation between each sparse unit and its adjacent non-sparse units in terms of terrain type and engineering stage, and calculate the unit correlation strength based on the logical correlation. Based on the unit association strength, the unit ecological vitality value that can be obtained by introducing new samples into the corresponding sparse unit is calculated. The higher the unit association strength, the greater the unit ecological vitality value. The set of unit ecological vitality values ​​of all grid cells constitutes the ecological vitality distribution map, and the unit ecological vitality value of non-sparse cells is zero.

[0037] In this embodiment, grid cells with a sample size below a preset threshold are marked as sparse cells, which is a preliminary screening mechanism based on statistics. This threshold is usually set according to the average density of the overall sample distribution or an empirical value. Its purpose is to quickly and objectively identify areas with insufficient sample coverage and missing information from a large number of grid cells as candidate targets for subsequent in-depth analysis. This is the initial step in distinguishing between regular areas and areas to be supplemented.

[0038] In this embodiment, the element association strength is a key quantitative indicator used to assess the potential value of sparse elements. Its calculation is not based on spatial geometric distance, but rather on engineering semantics and contextual logic. Specifically, it analyzes whether there are conventional construction process associations, risk transmission relationships, or scenario coexistence possibilities between the specific terrain and violation type represented by the sparse element and other terrains or engineering stages represented by adjacent non-sparse elements. For example, a sparse element representing slope-drainage ditch-siltation may have a high association strength with a dense element representing flat land-construction road-soil erosion, because runoff from slope construction often leads to downstream ditch siltation.

[0039] In this embodiment, the unit's ecological vitality value is an expected return indicator, calculated based on the unit's correlation strength, representing the supplementary value of the sparse unit. A higher correlation strength means that adding new samples to the sparse unit not only fills its own gaps but also forms an effective connection and mutual verification with the existing knowledge system of the surrounding dense areas, thereby significantly improving the systematicness and logical completeness of the entire power transmission and transformation project's environmental protection sample library. Therefore, its ecological vitality value is set higher to attract and prioritize its processing in subsequent processes.

[0040] In this embodiment, the ecological vitality distribution map is a spatial representation of the ecological vitality values ​​of individual units. This map clearly divides all grids into two categories: the vast majority of non-sparse units have a vitality value of zero, representing conventional areas; a few sparse units display varying levels of positive vitality values. These bright spots are precisely the high-potential target areas identified by the system as most worthy of targeted sample collection and annotation. It provides a direct data-driven navigation map for the proactive optimization and targeted construction of the environmental and water conservation sample bank for power transmission and transformation projects.

[0041] In this embodiment, the division of grid cells is the basis for spatial statistics. Similar to the calculation process of the ecological pressure distribution map, the entire three-dimensional classification space (consisting of terrain, targets, and violation types) is discretized into a series of cubes or grids of uniform size. This provides a unified statistical unit for independently calculating the number of samples and assessing sparsity within each subspace.

[0042] In this embodiment, identifying sparse cells is the first step in discovering knowledge blind spots. By presetting a sample size threshold (e.g., less than 2 samples), the system can automatically filter out all grid cells with weak coverage in the current sample pool. These sparse cells represent missing or incomplete areas in the current power transmission and transformation engineering environmental protection sample library under specific category combinations, and are potentially areas that need to be supplemented.

[0043] In this embodiment, logical correlation analysis is key to determining whether a sparse unit has high supplementary value. It does not simply calculate spatial proximity, but rather, based on prior knowledge of the engineering field, analyzes whether there are reasonable semantic or causal relationships between the sparse unit and the category combinations represented by adjacent non-sparse units. For example, a sparse unit of mountainous terrain-construction access road-soil erosion may have a strong logical correlation with a dense unit of flat land-spoil disposal site-improper stockpiling, because earth and stone excavated during mountainous construction are often transported to spoil disposal sites on flat land, and the two are closely related in the engineering chain.

[0044] In this embodiment, the unit's ecological vitality value is the core indicator for quantifying the priority of supplementing the sparse unit. Its value is directly determined by the unit's association strength; the stronger the association, the more helpful it is in connecting the knowledge graph and enhancing the model's understanding of related scenarios, thus resulting in higher vitality, i.e., higher expected gain. The system will prioritize supplementing samples with units that have high vitality values.

[0045] In this embodiment, the final generated ecological vitality distribution map is a global indicator map. In the map, all non-sparse units with sufficient samples have a vitality value of zero, while each sparse unit exhibits a different vitality value based on its correlation strength. This map intuitively reveals the specific dimensions in which the environmental protection sample library for power transmission and transformation projects has shortcomings, and which shortcomings most urgently need to be addressed, thus realizing the transformation from passive collection to proactive and targeted optimization.

[0046] To ensure that new samples can automatically adapt to competition strategies based on environmental pressure and maintain ecological balance within the database, it is proposed that the conversion of the first pressure value into a competition intensity coefficient is based on a first preset mapping relationship. The first preset mapping relationship defines that the larger the first pressure value, the larger the converted competition intensity coefficient.

[0047] In this embodiment, the first preset mapping relationship is the core transformation rule for realizing the environmental pressure-driven decision-making mechanism. It clearly establishes a positive correlation between the regional static pressure state (first pressure value) reflected by the ecological pressure distribution map and the intensity of the dynamic competition strategy (competition intensity coefficient) required when adding new samples to the database. Its function is to directly and automatically transform the macro-level regional pressure diagnosis results into micro-level database entry operation parameters, ensuring that the system strategy adapts to the environmental state in real time.

[0048] In this embodiment, the mapping relationship adopts a positive design where the higher the pressure value, the higher the competition intensity coefficient, which has clear engineering logic. When a new sample attempts to enter a high-pressure area, it indicates that the area already contains a large number of high-quality samples, and knowledge is approaching saturation. At this time, by increasing the competition intensity coefficient, the system will set stricter and more intense competition rules for entry into the database (e.g., reducing the competition radius to concentrate competition on direct competitors), thereby effectively filtering out new samples whose quality is insufficient to surpass the existing benchmark. This prevents low-quality samples from diluting the overall quality of the area, and is a key feedback control mechanism for maintaining regional quality stability and ecological balance.

[0049] To introduce a survival-of-the-fittest mechanism in high-density sample regions and achieve iterative updates of low-quality samples through competition, a competitive inclusion process is proposed, including: The competition radius is determined based on the competition intensity coefficient, and the competition radius is inversely proportional to the competition intensity coefficient. The competition region is defined as the spatial region within the competition radius centered on the classification spatial coordinates of the newly added samples. Within the competitive region, find the existing sample with the highest vitality score in the sample vitality profile and use it as the direct competing sample; Compare the viability scores of newly added samples with those of directly competing samples; If the viability score of the new sample is higher than that of the directly competing sample, the new sample is deemed to have won the competition and is accepted. The viability score of the directly competing sample is then reduced based on the competition intensity coefficient.

[0050] In this embodiment, the inverse relationship between the competition intensity coefficient and the competition radius is a sophisticated design that dynamically adjusts the scope and intensity of competition. In high-pressure regions (large competition intensity coefficients), the system sets a smaller competition radius, forcing new samples to engage in near-one-on-one peak battles with the strongest neighbor among existing samples (direct competitors). This significantly raises the entry threshold for new samples, ensuring that only samples significantly superior to the existing best can win, thereby strictly controlling the quality of high-density regions. Conversely, in low-pressure regions (small competition intensity coefficients), a larger competition radius means more relaxed competition, making it easier for new samples to find a place to survive and facilitating rapid filling of gaps.

[0051] In this embodiment, identifying the existing sample with the highest vitality score within the competitive region as the direct competitor and comparing it only with that sample is an efficient optimization strategy. This rule ensures that the competition is winner-takes-all and directly confronts the strongest. It avoids inefficient comparisons between new samples and a large number of ordinary samples, directly challenging the current champion in the region. This not only simplifies the decision-making logic but also ensures that the winner is necessarily the optimal individual in the region, thereby most effectively driving a one-way improvement in the quality of regional samples.

[0052] In this embodiment, the operation of reducing the vitality score of directly competing samples based on the competition intensity coefficient is a key closed loop for completing quality iteration. If a new sample wins, the challenged directly competing sample (the original champion) is not immediately deleted, but its vitality score is weakened proportionally according to the competition intensity coefficient. This both makes way for the inclusion of new samples and simulates the decay of the relative value of old knowledge in the knowledge base. In high-pressure areas, the weakening effect of winning is stronger, accelerating the elimination process of the original high-quality samples; in low-pressure areas, the weakening effect is weaker, preserving more historical knowledge. This mechanism together ensures that the samples in the database continue to evolve towards higher quality.

[0053] To stimulate synergistic gain effects among samples in sparse or complementary regions, thereby enhancing the diversity and systematic value of the overall power transmission and transformation engineering environmental and water conservation sample library, a synergistic gain process is proposed, including: The coordination radius is determined based on the coordination strength coefficient, which is directly proportional to the coordination radius. Using the classification spatial coordinates of the newly added samples as the center, a spatial region within the collaborative radius is defined as the collaborative region; Within the collaborative region, existing samples that are complementary to the new samples in terms of environmental protection and water conservation target types or engineering violation types are identified and used as collaborative gain samples. If a collaborative gain sample is found, it is determined that the new sample triggers collaboration, the new sample is accepted, and the vitality score of the new sample and all collaborative gain samples is increased according to the collaboration strength coefficient.

[0054] In this embodiment, the direct proportionality between the synergy strength coefficient and the synergy radius is the core strategy for guiding and amplifying the symbiotic effect. In highly active regions (with large synergy strength coefficients), the system sets a larger synergy radius to search for potential synergistic samples as widely as possible, stimulating the maximum range of complementary effects and thus rapidly strengthening regions with weak knowledge but strong connections. This design encourages new samples to act as catalysts for connecting and activating a network of interconnected knowledge, rather than existing in isolation.

[0055] In this embodiment, finding synergistic gain samples based on complementary relationships between environmental protection target types or engineering violation types is a rule that defines the essence of synergy. Here, complementarity is not simply a difference in category, but rather a relationship of mutual association, mutual explanation, or constituting a complete process in engineering logic or scenario. For example, if a new sample is a tower base area-vegetation destruction type, its synergistic gain samples might be a tower base area-temporary cover or a construction access road-soil erosion. Together, they describe a more complete scenario chain from tower base construction, from destruction to temporary protection, and the potential associated impacts. This search mechanism aims to construct a semantic network between samples, rather than simply piling them up in isolation.

[0056] In this embodiment, increasing the vitality score of newly added samples and all synergistic gain samples based on the synergy strength coefficient is an incentive feedback mechanism to achieve shared value amplification. Once synergy is triggered, not only are new samples accepted, but the vitality scores of all related existing samples are also increased. In high-vitality areas (where the synergy strength coefficient is large), this increase is even greater. This is equivalent to the system providing additional rewards for sample combinations that appear in groups and can form a knowledge loop, thereby encouraging and solidifying this systematic and diverse knowledge structure, making the overall value of the power transmission and transformation engineering environmental protection sample library greater than the simple sum of individual samples.

[0057] To simulate the knowledge aging process and automatically reduce the weight of old or low-sensitivity samples, ensuring the timeliness and dynamism of the environmental protection sample database for power transmission and transformation projects, a vitality decay operation is proposed to follow an exponential decay model, in which the vitality score in the sample vitality archive decreases exponentially with time interval. The decay rate factor is set according to the environmental sensitivity level associated with the engineering violation type labeled in the sample. The higher the environmental sensitivity level, the smaller the decay rate factor, and the slower the vitality score decreases.

[0058] In this embodiment, the exponential decay model is the core mathematical tool for simulating the natural dissipation of sample knowledge value over time. This model sets the sample's vitality score to decrease non-linearly and uniformly, but rather to shrink by a fixed percentage of the current score after each fixed period. This means that samples with high initial values ​​or those that have been recently updated experience a greater absolute decay, quickly widening the gap with older samples; while low-score samples, after multiple decays, will experience a more gradual decline. This accurately simulates the forgetting curve of knowledge from fresh to outdated, and then to a point of stabilization.

[0059] In this embodiment, setting the decay rate factor based on the environmental sensitivity level associated with the type of engineering violation reflects the introduction of domain prior knowledge for refined management. For samples involving highly sensitive violations such as permanent ecological damage and significant environmental risks (e.g., damage to water sources or habitats of rare species), the knowledge has long-term reference value and is therefore assigned a smaller decay rate factor, allowing its vitality to decline slowly and enabling it to remain in the database for a long time. Conversely, for samples of temporary violations with low environmental impact (e.g., short-term construction dust), their timeliness is high, so a larger decay rate factor is assigned, causing its value to decay rapidly, making room for newer samples. This differentiated decay ensures a balance between the timeliness and the durability of key knowledge in the power transmission and transformation engineering environmental protection sample database.

[0060] To implement differentiated and prioritized elimination strategies based on varying regional competitive pressures and efficiently release resources occupied by low-value samples, the following sample elimination operations are proposed: Different survival thresholds are set for grid cells with different ecological pressure values ​​in the ecological pressure distribution map. The higher the ecological pressure value of a cell, the higher the corresponding survival threshold. Traverse each grid cell and compare the vitality score in the vitality profile of the samples in that cell with the cell survival threshold. Samples with a vitality score lower than the survival threshold of their respective grid cells are marked as samples to be eliminated. When eliminating samples, they are removed from the power transmission and transformation project environmental protection sample pool in order of increasing sample entropy value, with samples of low entropy value being eliminated first.

[0061] In this embodiment, dynamically setting different survival thresholds for different units based on their ecological pressure values ​​is the core strategy for differentiated elimination. The design logic is that high-pressure areas typically have dense sample densities and high overall quality. To maintain this high quality standard and prevent overcrowding, the system sets a higher survival threshold (unit survival threshold), ensuring that only highly resilient samples survive. Conversely, in low-pressure areas, samples are scarce or of average quality, resulting in a lower survival threshold to retain more samples and encourage diversity. This simulates the natural law in the ecological environment where resource-scarce, highly competitive areas place higher demands on individual survival capabilities.

[0062] In this embodiment, samples with a vitality score below the survival threshold of their respective grid cells are marked as candidates for elimination, representing a preliminary screening based on a single quantitative indicator. This step quickly identifies all samples that are unqualified under the current regional competition criteria, regardless of their absolute scores. For example, a sample with a vitality score of 70 would be marked for elimination in a high-pressure area (threshold set at 80), while the same sample might survive in a low-pressure area (threshold set at 60). This ensures that the elimination criteria are adapted to the regional environment.

[0063] In this embodiment, prioritizing the elimination of low-entropy samples according to their entropy values ​​(from lowest to highest) is a refined prioritization strategy implemented after the initial screening. Low-entropy samples mean that their characteristics are very close to the central pattern of their category, i.e., ordinary or common samples. Prioritizing the elimination of these samples can, while cleaning up low-value data, preserve as many unique and rare high-entropy samples as possible, thereby effectively maintaining the overall diversity and coverage of the power transmission and transformation engineering environmental protection sample library, and avoiding the loss of valuable marginal cases or rare morphological samples due to elimination.

[0064] To transform the dynamic update process into interpretable and executable optimization suggestions, and to achieve closed-loop monitoring and proactive optimization of the environmental and water conservation sample database for power transmission and transformation projects, the following additional measures are proposed: Obtain the current ecological pressure distribution map and ecological vitality distribution map of the environmental protection and water conservation sample database for power transmission and transformation projects; In the ecological pressure distribution map, areas where the unit ecological pressure value is consistently higher than the high-pressure warning threshold are marked as high-pressure bottleneck areas. In the ecological vitality distribution map, areas where the unit ecological vitality value is consistently higher than the high potential warning threshold are marked as high potential sparse areas. Analyze the sample vitality profiles of samples within the high-pressure bottleneck area, and statistically analyze the proportion of samples with low vitality scores and the proportion of samples with high entropy values. Analyze the engineering semantics associated with high-potential sparse areas to determine the expected terrain type, environmental protection target type, and engineering violation type of the samples to be supplemented; Generate an optimization report for the environmental protection and water conservation sample library of power transmission and transformation projects. The optimization report includes the location information of high-voltage bottleneck areas, the sample quality analysis results within the areas, the location information of high-potential sparse areas, and the expected characteristic description of samples to be supplemented. Based on the optimization report, optimization suggestions for the environmental protection and water conservation sample library of power transmission and transformation projects are output. The optimization suggestions include suggestions for cleaning up samples in high-voltage bottleneck areas and suggestions for targeted sample collection and labeling in high-potential sparse areas.

[0065] In this embodiment, the identification of high-voltage bottleneck areas and high-potential-sparse areas is a crucial step in achieving proactive diagnosis. High-voltage bottleneck areas refer to regions that have long been characterized by high sample density and high competition intensity. These areas may have accumulated a large number of redundant or homogeneous samples, representing potential bottlenecks that lead to resource waste and stagnant updates. High-potential-sparse areas, on the other hand, are regions that have long been identified as having scarce samples but extremely strong correlations. They represent critical gaps in the coverage of the environmental and water conservation sample bank for power transmission and transformation projects. By setting early warning thresholds and continuous monitoring, the system can automatically and accurately identify these two types of typical problem areas requiring focused intervention.

[0066] In this embodiment, the sample quality analysis of the high-pressure bottleneck area aims to provide a basis for precise cleanup. Analyzing the proportion of samples with low viability scores can quantitatively assess the overall health and redundancy of the samples in this area; analyzing the proportion of samples with high entropy values ​​helps to determine whether there will be excessive loss of diversity during the potential elimination process. This quantitative analysis makes the cleanup recommendations for the high-pressure area no longer a vague statement of needing cleanup, but rather specifically points to how many low-quality samples should be cleaned up and which unique samples should be prioritized for retention.

[0067] In this embodiment, engineering semantic analysis of high-potential sparse areas is the core step in transforming data gaps into specific data collection tasks. The system not only identifies where samples are missing, but more importantly, it infers what types of samples should be added by analyzing the logical relationships between these gaps and adjacent dense areas. For example, the system might determine that a high-potential sparse area needs to be supplemented with samples from hilly areas, areas involving temporary soil dumps, or areas showing early signs of soil erosion. This makes subsequent optimization suggestions clearly actionable.

[0068] In this embodiment, the final generated optimization report and optimization suggestions complete a closed loop from data perception to management decision-making. The optimization report is an interpretable diagnostic manual for managers, clearly presenting where the problems are, what the problems are, and the direction of solutions. The optimization suggestions based on the report output serve as direct action guidelines for the maintainers or data annotation teams of the power transmission and transformation project environmental and water conservation sample library. For example, it suggests conducting sample quality verification and cleanup for Class A high-voltage areas and organizing targeted remote sensing photography and annotation for Class B high-potential areas. This achieves a fundamental shift in the maintenance of the power transmission and transformation project environmental and water conservation sample library from passive response to proactive planning and data-driven approaches.

[0069] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of this invention and its equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for classifying and dynamically updating remote sensing image samples related to water conservation in power transmission and transformation projects, characterized in that... include: Construct a sample library of environmental protection and water conservation for power transmission and transformation projects, and create and maintain a sample lifespan profile for each sample in the sample library; The classification space coordinates of each sample are determined according to the preset three-dimensional classification rules, which include the terrain type dimension, the environmental protection and water conservation target dimension, and the engineering violation type dimension. Based on the classification spatial coordinates and sample vitality profiles of all samples, ecological pressure distribution maps and ecological vitality distribution maps of the environmental protection and water conservation sample bank for power transmission and transformation projects are calculated and generated. Receive new remote sensing image samples and generate sample vitality profiles and classification spatial coordinates for the new samples; Based on the classification spatial coordinates of the new samples, the first pressure value is extracted from the ecological pressure distribution map, and the first vitality value is extracted from the ecological vitality distribution map. Convert the first pressure value into a competition intensity coefficient and the first vitality value into a synergy intensity coefficient; Based on the competition intensity coefficient, the competitive entry process is executed: determine whether the new sample wins the competition, and if so, accept the new sample and adjust the sample lifespan profile of the competing samples in the database. Based on the collaboration strength coefficient, the collaboration gain process is executed: determine whether the new sample triggers collaboration, and if so, accept the new sample and improve the sample life file of the new sample and the collaborative samples in the library. Perform a vitality decay operation on the sample vitality profiles of all samples periodically; Based on the weakened sample viability profile, a sample elimination operation is performed to remove samples with weak viability, and the ecological pressure distribution map and ecological vitality distribution map are recalculated. Output the updated sample library for environmental protection and water conservation in power transmission and transformation projects.

2. The method for classifying and dynamically updating remote sensing image samples for water conservation in power transmission and transformation projects according to claim 1, characterized in that, The sample vitality profile consists of three sub-items: vitality score, sample entropy value, and sample popularity. The vitality score is calculated based on the clarity of the sample image, the consistency of the annotation, and the typicality of the visual features. The sample entropy value is calculated based on the distance between the visual features of a sample and the cluster center of similar sample features. The greater the distance, the higher the sample entropy value. The sample popularity record shows the number of times the sample has been successfully retrieved or accessed in history.

3. The method for classifying and dynamically updating remote sensing image samples for water conservation in power transmission and transformation projects according to claim 1, characterized in that, Calculate the ecological pressure distribution map, including: The three-dimensional classification space is divided into multiple uniform grid units; For each grid cell, count the number of samples within the cell and calculate the cell sample density index based on the number of samples. For each grid cell, the average of the vitality scores of all samples within the cell is calculated, and the cell quality competition index is calculated based on this average. The unit sample density index and the unit quality competition index of each grid cell are weighted and summed to obtain the unit ecological pressure value of the corresponding grid cell. The set of ecological pressure values ​​for all grid cells constitutes an ecological pressure distribution map.

4. The method for classifying and dynamically updating remote sensing image samples for water conservation in power transmission and transformation projects according to claim 1, characterized in that, Calculating the distribution map of ecological vitality includes: The three-dimensional classification space is divided into multiple uniform grid units; Set a threshold for the number of samples, and mark grid cells with a sample number lower than the threshold as sparse cells; Analyze the logical correlation between each sparse unit and its adjacent non-sparse units in terms of terrain type and engineering stage, and calculate the unit correlation strength based on the logical correlation. Based on the unit association strength, the unit ecological vitality value that can be obtained by introducing new samples into the corresponding sparse unit is calculated. The higher the unit association strength, the greater the unit ecological vitality value. The set of unit ecological vitality values ​​of all grid cells constitutes the ecological vitality distribution map, and the unit ecological vitality value of non-sparse cells is zero.

5. The method for classifying and dynamically updating remote sensing image samples for water conservation in power transmission and transformation projects according to claim 1, characterized in that, The conversion of the first pressure value into a competition intensity coefficient is based on a first preset mapping relationship. The first preset mapping relationship defines that the larger the first pressure value, the larger the competition intensity coefficient obtained.

6. The method for classifying and dynamically updating remote sensing image samples for water conservation in power transmission and transformation projects according to claim 1, characterized in that, The competitive warehousing process includes: The competition radius is determined based on the competition intensity coefficient, and the competition radius is inversely proportional to the competition intensity coefficient. The competition region is defined as the spatial region within the competition radius centered on the classification spatial coordinates of the newly added samples. Within the competitive region, find the existing sample with the highest vitality score in the sample vitality profile and use it as the direct competing sample; Compare the viability scores of newly added samples with those of directly competing samples; If the viability score of the new sample is higher than that of the directly competing sample, the new sample is deemed to have won the competition and is accepted. The viability score of the directly competing sample is then reduced based on the competition intensity coefficient.

7. The method for classifying and dynamically updating remote sensing image samples for water conservation in power transmission and transformation projects according to claim 1, characterized in that, The collaborative gain process includes: The coordination radius is determined based on the coordination strength coefficient, which is directly proportional to the coordination radius. Using the classification spatial coordinates of the newly added samples as the center, a spatial region within the collaborative radius is defined as the collaborative region; Within the collaborative region, existing samples that are complementary to the new samples in terms of environmental protection and water conservation target types or engineering violation types are identified and used as collaborative gain samples. If a collaborative gain sample is found, it is determined that the new sample triggers collaboration, the new sample is accepted, and the vitality score of the new sample and all collaborative gain samples is increased according to the collaboration strength coefficient.

8. The method for classifying and dynamically updating remote sensing image samples for water conservation in power transmission and transformation projects according to claim 1, characterized in that, The vitality decay operation follows an exponential decay model, where the vitality score in the sample vitality file decreases exponentially with time intervals. The decay rate factor is set according to the environmental sensitivity level associated with the engineering violation type labeled in the sample. The higher the environmental sensitivity level, the smaller the decay rate factor, and the slower the vitality score decreases.

9. The method for classifying and dynamically updating remote sensing image samples for water conservation in power transmission and transformation projects according to claim 1, characterized in that, The sample elimination process includes: Different survival thresholds are set for grid cells with different ecological pressure values ​​in the ecological pressure distribution map. The higher the ecological pressure value of a cell, the higher the corresponding survival threshold. Traverse each grid cell and compare the vitality score in the vitality profile of the samples in that cell with the cell survival threshold. Samples with a vitality score lower than the survival threshold of their respective grid cells are marked as samples to be eliminated. When eliminating samples, they are removed from the power transmission and transformation project environmental protection sample pool in order of increasing sample entropy value, with samples of low entropy value being eliminated first.

10. The method for classifying and dynamically updating remote sensing image samples for water conservation in power transmission and transformation projects according to claim 1, characterized in that, Also includes: Obtain the current ecological pressure distribution map and ecological vitality distribution map of the environmental protection and water conservation sample database for power transmission and transformation projects; In the ecological pressure distribution map, areas where the unit ecological pressure value is consistently higher than the high-pressure warning threshold are marked as high-pressure bottleneck areas. In the ecological vitality distribution map, areas where the unit ecological vitality value is consistently higher than the high potential warning threshold are marked as high potential sparse areas. Analyze the sample vitality profiles of samples within the high-pressure bottleneck area, and statistically analyze the proportion of samples with low vitality scores and the proportion of samples with high entropy values. Analyze the engineering semantics associated with high-potential sparse areas to determine the expected terrain type, environmental protection target type, and engineering violation type of the samples to be supplemented; Generate an optimization report for the environmental protection and water conservation sample library of power transmission and transformation projects. The optimization report includes the location information of high-voltage bottleneck areas, the sample quality analysis results within the areas, the location information of high-potential sparse areas, and the expected characteristic description of samples to be supplemented. Based on the optimization report, optimization suggestions for the environmental protection and water conservation sample library of power transmission and transformation projects are output. The optimization suggestions include suggestions for cleaning up samples in high-voltage bottleneck areas and suggestions for targeted sample collection and labeling in high-potential sparse areas.

Images