Intelligent forest pest risk level assessment method based on multi-source sensing data

Abstract

This invention discloses an intelligent assessment method for forest pest risk levels based on multi-source sensor data, comprising: acquiring multi-source sensor data, completing preprocessing, and forming a standardized dataset; processing multi-temporal remote sensing images, improving SNUNet-CD to detect changes, and generating canopy gradient response sequences; constructing microclimate-based canopy sequences, and extracting phase closed-loop transition features through hysteresis map encoding; constructing a propagation network, sequentially implementing wind direction, host, topography, and phenology gating to form propagation paths and critical indices; calculating initial risk values, performing perturbation inversion of risk-causing factors, determining minimum risk conditions, and obtaining stability; fusing various indicators to obtain a comprehensive risk index, correcting it, classifying risk levels, and outputting assessment results. This invention achieves early identification, accurate assessment, and interpretable classification of forest pest risks through deep coupling modeling of multi-source sensor data and pest occurrence mechanisms.

Description

Technical Field

[0001] This invention relates to the field of forestry disaster monitoring and intelligent assessment technology, and in particular to an intelligent assessment method for forest pest risk levels based on multi-source sensor data. Background Technology

[0002] Forest pests are a significant factor affecting the stability of forest ecosystems and the security of forestry resources. Their occurrence is characterized by high concealment, rapid spread, and wide-ranging damage. With the development of remote sensing, IoT sensing, and data analysis technologies, various methods for monitoring and assessing forest pest risks based on single or multi-source data have been proposed. For example, identifying affected areas by extracting vegetation index changes from remote sensing images, analyzing pest occurrence conditions by deploying sensors in the forest to obtain environmental data such as temperature, humidity, and rainfall, or combining data from traps to statistically analyze pest populations, and some methods employ machine learning or statistical models to fuse multi-source data to predict the probability or risk level of forest pest occurrence.

[0003] However, existing technologies still have significant shortcomings in practical applications. Most methods rely on simple fusion or weighted calculations of multi-source data for risk assessment, lacking in-depth modeling of pest occurrence mechanisms. They struggle to effectively couple the intrinsic relationships between environmental triggers, pest activity, and vegetation symptoms, resulting in assessments that often remain at a static characteristic level, failing to reflect the dynamic evolution of pest occurrence from triggering to development to symptom onset. Existing methods generally focus on single-phase or limited-phase data analysis, lacking effective characterization of lag relationships between multiple time-series data, making it difficult to identify early stages of pests and affecting the foresight and accuracy of early warnings. Spatially, most methods analyze only simple neighborhood relationships, lacking the ability to model the directional spread paths of pests under the combined influence of wind direction, topography, and host distribution, making it difficult to accurately describe the spread trends of pests.

[0004] Existing risk assessment methods also have shortcomings in terms of the reliability of results. Due to the lack of inversion analysis of key risk factors and verification mechanisms for the stability of risk results, some methods are prone to misjudgment when faced with environmental anomalies or interference from non-pest factors, resulting in insufficient credibility of high-risk results and thus affecting the scientific nature of prevention and control decisions.

[0005] Therefore, how to provide an intelligent assessment method for forest pest risk levels based on multi-source sensor data is a problem that urgently needs to be solved by those skilled in the art. Summary of the Invention

[0006] One objective of this invention is to propose an intelligent assessment method for forest pest risk levels based on multi-source sensor data. This invention fully utilizes the improved SNUNet-CD model, cross-modal hysteresis map modeling technology, and spatial seepage propagation modeling technology. By improving the SNUNet-CD model, it achieves the extraction of canopy gradient response from multi-temporal remote sensing images, constructs a closed-loop identification mechanism for pest occurrence hysteresis and a mechanism for inferring directional propagation paths, and corrects the risk results by combining the inversion of risk-causing factors. It describes the dynamic assessment process of forest pests from induction, symptom onset to spread under multi-source data-driven conditions, and has the advantages of strong early identification capability, accurate spatial propagation characterization, high reliability of assessment results, and strong interpretability.

[0007] The intelligent assessment method for forest pest risk levels based on multi-source sensor data according to embodiments of the present invention includes: Acquire multi-source sensor data of the target forest area, preprocess the multi-source sensor data, and obtain standardized datasets corresponding to each risk assessment unit; For remote sensing image data acquired in consecutive time phases for the same risk assessment unit, an improved SNUNet-CD model is used to detect changes in adjacent time phases, resulting in a canopy change mask and change intensity map. Based on the time-series cumulative mapping, a canopy gradual response map is generated, forming a canopy gradual response sequence. Based on standardized datasets, microclimate-induced sequences and insect activity sequences were constructed. Combined with canopy gradual response sequences, cross-modal hysteresis map encoding was performed to extract the phase shift relationship, closed-loop evolution trajectory and stage transition characteristics between microclimate-induced, insect activity and canopy gradual response, and generate insect hysteresis closed-loop index. Based on the spatial propagation influence factors between risk assessment units, an on / off anisotropic seepage propagation map is constructed, which includes wind direction consistent gate, host continuity gate, topographic diversion gate and phenological synchronous gate, and the pest propagation path and seepage critical index are calculated. Based on the pest lag closed-loop index, seepage critical index and host susceptibility characteristics, the initial pest risk value of each risk assessment unit is calculated. For risk assessment units with risk values ​​higher than the preset threshold, the perturbation inversion of risk-causing factors is performed to determine the minimum set of risk-causing conditions and calculate the risk stability index. Based on the initial pest risk value, pest lag closed-loop index, seepage critical index, and risk stability index, the risk level of each risk assessment unit is classified, and the corresponding forest pest risk level results and information on the main risk factors are output.

[0008] Optionally, the multi-source sensor data includes remote sensing image data, forest environment sensor data, insect monitoring data, meteorological data, forest stand structure data, and topographic data.

[0009] Optionally, the preprocessing of multi-source sensor data includes outlier removal and missing value completion for various types of data, time alignment of data from different sources according to a unified time scale, spatial mapping of data with different spatial resolutions according to risk assessment units, coordinate system unification of all data, and normalization of various feature data.

[0010] Optionally, obtaining the canopy change mask and change intensity map, and generating a canopy gradient response map based on time-series cumulative mapping to form a canopy gradient response sequence includes: The remote sensing image data of the same risk assessment unit are acquired in consecutive time phases. Geometric correction, radiometric correction, spatial registration and cropping are performed on the remote sensing images of each time phase, and adjacent time phase image pairs are constructed in chronological order. An improved SNUNet-CD model is constructed, which includes a temporal difference guided coding layer, a gradient enhancement decoding layer, and a response constraint output layer. The temporal difference guided coding layer extracts isomorphic features from adjacent temporal images and generates temporal difference features. The gradient enhancement decoding layer fuses temporal difference features and canopy texture features at different scales step by step. The response constraint output layer outputs the fused features in a dual-branch manner. The SNUNet-CD model is improved by inputting adjacent temporal image pairs. The corresponding layer features of adjacent temporal images are extracted using the time difference guided coding layer. Temporal difference guided feature maps are constructed between each corresponding layer. The temporal difference guided feature maps of each layer are then input into the gradient enhancement decoding layer for step-by-step upsampling and skip connection fusion to obtain the canopy change characterization feature map. The canopy change feature map is input into the response constraint output layer. One output branch performs pixel-level classification of the change area to obtain the canopy change mask, and the other output branch performs hierarchical characterization of the degree of change in the change area to obtain the change intensity map. The canopy change mask and the change intensity map are mapped according to the risk assessment unit to obtain the change range and change intensity results corresponding to each risk assessment unit. By performing time-series cumulative mapping on the canopy change mask and change intensity map corresponding to each adjacent time phase, the change range, change intensity and change location from the previous time phase to the current time phase are superimposed and correlated to generate a canopy gradual change response map reflecting the process of canopy anomaly from initial appearance to continuous expansion. Based on the canopy gradual change response map, the initial appearance time, duration, expansion rate, plaque connectivity growth rate and directional migration characteristics of the canopy anomaly are extracted and correlated and encoded in time sequence to form a canopy gradual change response sequence corresponding to each risk assessment unit.

[0011] Optionally, the generation of the pest hysteresis closed-loop index includes: Based on forest environment sensor data, microclimate features related to pest induction were extracted, and a microclimate induction sequence was constructed according to a unified time scale. Based on pest monitoring data, pest population change features and pest source activity features were extracted, and a pest source activity sequence was constructed according to a unified time scale. The canopy gradual response sequence was called as the symptomatic response status of each risk assessment unit in a continuous time series. The microclimate-induced sequence, insect activity sequence, and canopy gradual response sequence were synchronized and aligned. Ternary state units were constructed according to their temporal order. Each ternary state unit includes the environmental induced state, insect activity state, and symptomatic response state at the same time. Based on all ternary state units, cross-modal hysteresis maps corresponding to each risk assessment unit were generated. Hysteresis stage encoding was performed on the cross-modal hysteresis map. The state changes in the microclimate-induced sequence were used as the initial driver, the state changes in the insect source activity sequence were used as the relay, and the state changes in the canopy gradual response sequence were used as the terminal response. The phase relationship, closed-loop characteristics and stage transition characteristics between microclimate induction, insect source activity and canopy gradual response were extracted. Closed-loop constraint encoding is performed on the cross-modal hysteresis map. Following the order of microclimate induction first, insect source activity followed, and canopy gradual change response manifested later, state chains that do not meet the order are suppressed, while state chains that meet the order and are continuously maintained are enhanced. Based on the enhanced state chains, the hysteresis closed-loop strength, closed-loop duration, and closed-loop stability are extracted to form a hysteresis closed-loop feature set. Based on the hysteresis closed-loop feature set, the phase relationship consistency, closed-loop continuity and stage transition integrity of each risk assessment unit are jointly encoded to generate the pest hysteresis closed-loop index.

[0012] Optionally, the calculation of pest transmission pathways and seepage critical index includes: Using risk assessment units as graph nodes, initial connecting edges are established according to the boundary contact relationship and spatial position relationship between adjacent risk assessment units to form a basic propagation network between risk assessment units, and each connecting edge is assigned a propagation direction identifier. Based on wind direction and speed data, wind direction consistency gates are used to determine the wind direction of each connecting edge in the basic propagation network. The prevailing wind direction is matched with the propagation direction of the connecting edge. The matching connecting edge is given an open state, and the unmatched connecting edge is given a closed state or a suppressed state, forming the first propagation edge set with directional propagation constraints. Based on stand structure data, host continuity gate determination is performed on the first propagation edge set. According to the consistency of host tree species, the continuity of host distribution, and the similarity of stand age between adjacent risk assessment units, the connection edge that meets the continuous host condition is given an enhanced state, and the connection edge that does not meet the continuous host condition is given a weakened state, thus forming a second propagation edge set with host transmission constraints. Based on topographic data, the second propagation edge set is determined by topographic guidance gate. According to the slope aspect connection relationship, slope transition relationship and elevation guidance relationship between adjacent risk assessment units, the connecting edge that is conducive to the spread of pests along the topographic direction is identified. The connecting edge is given a guidance enhancement state, and the connecting edge that is not conducive to the spread is given a blocking state, thus forming a third propagation edge set with topographic guidance constraints. Based on the pest hysteresis closed-loop index, the third propagation edge set is used to determine the phenological synchronization gate. According to whether the hysteresis closed-loop state of adjacent risk assessment units is in an adjacent stage or a continuous stage, the connecting edge with synchronous propagation conditions is identified. The connecting edge that simultaneously satisfies the wind direction consistent gate, host continuous gate, topographic diversion gate and phenological synchronization gate is determined as an effective propagation edge. According to the continuous connection relationship of the effective propagation edge, the switch-type anisotropic seepage propagation map is constructed. Among them, the effective propagation edges that are continuously open in the same time period are combined to form a stage propagation channel. The stage propagation channel that remains open in adjacent time periods is determined as a stable propagation channel. Based on the switch-type anisotropic seepage propagation map, the effective propagation edges from risk assessment units with insect activity indices higher than the preset insect threshold to other risk assessment units are traversed step by step to determine the insect propagation path. The seepage critical index is calculated based on the number of stable propagation channels, the range of risk assessment units covered by the stable propagation channels, and the degree of continuity of the propagation path.

[0013] Optionally, the calculation of the initial pest risk value for each risk assessment unit, the inversion of risk-causing factor perturbation for risk assessment units exceeding a preset threshold, the determination of the minimum risk-causing condition set, and the calculation of the risk stability index include: Based on the pest lag closed-loop index, seepage critical index, and host susceptibility characteristics of each risk assessment unit, the initial pest risk value of each risk assessment unit is calculated according to the preset risk fusion rules, and the risk assessment units whose initial pest risk value reaches the preset risk threshold are selected as the inversion units. For each unit to be inverted, a set of risk-causing disturbances is constructed. The set of risk-causing disturbances includes hysteresis closed-loop disturbances, propagation channel disturbances, and host susceptibility disturbances. Candidate disturbance paths are generated by taking the three types of disturbances in the order of single-factor disturbances, two-factor linkage disturbances, and three-factor linkage disturbances. Each candidate disturbance path is subjected to a step-by-step disturbance inversion of risk factors. The disturbances are applied to the unit to be inverted in order from low disturbance level to high disturbance level. After each level of disturbance, the corresponding risk value is recalculated. The linkage changes of the hysteresis closed loop state, the propagation channel state and the host susceptibility state are recorded during the risk value decrease process, forming a risk factor backtracking chain corresponding to each unit to be inverted. In the risk factor backtracking chain, candidate disturbance paths that cause the risk value of the unit to be inverted to drop below the preset risk threshold for the first time are selected, and the candidate disturbance path with the lowest disturbance level, the fewest number of risk factors involved and the shortest risk backtracking path is determined as the minimum risk condition set. The risk stability index is calculated based on the minimum set of risk-causing conditions. The risk stability index is jointly determined by the disturbance level, the number of risk-causing factors, the risk retreat magnitude after the disturbance, and the continuity of the risk-causing factor retreat chain in the minimum set of risk-causing conditions.

[0014] Optionally, the output of the corresponding forest pest risk level result and main risk factor information includes: The initial pest risk value, pest lag loop index, seepage critical index and risk stability index of each risk assessment unit are obtained. The indicators are processed to a uniform scale and the indicators of different dimensions are converted into comparable standardized indicator values. Based on the standardized initial pest risk value, pest lag closed-loop index, seepage critical index and risk stability index, the indicators are weighted and integrated according to the preset weights to obtain the comprehensive risk index of each risk assessment unit. A risk stability threshold is set, and the comprehensive risk index is corrected based on the comparison between the risk stability index and the risk stability threshold. Specifically, when the risk stability index reaches or exceeds the risk stability threshold, the comprehensive risk index remains unchanged. When the risk stability index is lower than the risk stability threshold, the comprehensive risk index is adjusted downward to obtain the corrected risk index. Multiple risk level classification threshold ranges are set, and the corrected risk index of each risk assessment unit is compared with the threshold range. The corresponding risk level is determined according to the threshold range in which the corrected risk index is located. Based on the risk level of each risk assessment unit and the corresponding pest lag closed-loop index, seepage critical index and risk stability index, the main risk factors are determined, and the risk level results and main risk factor information of each risk assessment unit are output.

[0015] The beneficial effects of this invention are: This invention achieves a shift from data-driven to mechanism-driven approaches by structurally coupling multi-source sensor data with the mechanisms of forest pest occurrence. It breaks through the limitations of existing assessment methods that primarily rely on static feature fusion. By introducing a multi-temporal canopy gradient response extraction method based on an improved SNUNet-CD model, the continuous variation information implicit in remote sensing images is transformed into a gradient response sequence that can be used for analysis. This effectively identifies the entire process of forest pests from initial induction to symptomatic expansion, improving the early identification capability and timeliness of pest assessment.

[0016] This invention constructs a cross-modal hysteresis map coding mechanism to unify the modeling of three heterogeneous data types: microclimate-induced, insect source activity, and canopy gradual response. This accurately depicts the temporal sequence and closed-loop evolution process among these three factors, overcoming the limitation of existing technologies in identifying the dynamic chain of pest induction-development-symptom onset. By introducing a switch-type anisotropic seepage propagation map, and comprehensively constraining modeling with multiple factors such as wind direction, topography, and host continuity, it achieves the identification of directional propagation paths of pests in space. This expands risk assessment from single-point analysis to regional propagation analysis, improving the accuracy of spatial prediction.

[0017] This invention utilizes a risk factor perturbation inversion mechanism to identify key factor combinations influencing risk formation, thereby determining the minimum risk-causing condition set and risk stability. This allows for the correction and verification of initial risk results, effectively reducing false alarm probability and improving the credibility and interpretability of risk assessment results. While ensuring feasibility, this invention achieves a comprehensive improvement in forest pest risk assessment across time and space dimensions, as well as result reliability, providing more scientific and reliable technical support for precise control of forest pests. Attached Figure Description

[0018] 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 The flowchart shows the intelligent assessment method for forest pest risk levels based on multi-source sensor data proposed in this invention. Figure 2 This is a flowchart of the canopy gradient response extraction process based on the improved SNUNet-CD model in the intelligent assessment method for forest pest risk levels based on multi-source sensor data proposed in this invention. Detailed Implementation

[0019] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0020] refer to Figure 1 and Figure 2 A smart assessment method for forest pest risk levels based on multi-source sensor data includes: Acquire multi-source sensor data of the target forest area, preprocess the multi-source sensor data, and obtain standardized datasets corresponding to each risk assessment unit; For remote sensing image data acquired in consecutive time phases for the same risk assessment unit, an improved SNUNet-CD model is used to detect changes in adjacent time phases, resulting in a canopy change mask and change intensity map. Based on the time-series cumulative mapping, a canopy gradual response map is generated, forming a canopy gradual response sequence. Based on standardized datasets, microclimate-induced sequences and insect activity sequences were constructed. Combined with canopy gradual response sequences, cross-modal hysteresis map encoding was performed to extract the phase shift relationship, closed-loop evolution trajectory and stage transition characteristics between microclimate-induced, insect activity and canopy gradual response, and generate insect hysteresis closed-loop index. Based on the spatial propagation influence factors between risk assessment units, an on / off anisotropic seepage propagation map is constructed, which includes wind direction consistent gate, host continuity gate, topographic diversion gate and phenological synchronous gate, and the pest propagation path and seepage critical index are calculated. Based on the pest lag closed-loop index, seepage critical index and host susceptibility characteristics, the initial pest risk value of each risk assessment unit is calculated. For risk assessment units with risk values ​​higher than the preset threshold, the perturbation inversion of risk-causing factors is performed to determine the minimum set of risk-causing conditions and calculate the risk stability index. Based on the initial pest risk value, pest lag closed-loop index, seepage critical index, and risk stability index, the risk level of each risk assessment unit is classified, and the corresponding forest pest risk level results and information on the main risk factors are output.

[0021] In this embodiment, the multi-source sensor data includes remote sensing image data, forest environment sensor data, insect monitoring data, meteorological data, forest stand structure data, and topographic data.

[0022] In this embodiment, the preprocessing of multi-source sensor data includes outlier removal and missing value completion for various types of data, time alignment of data from different sources according to a unified time scale, spatial mapping of data with different spatial resolutions according to risk assessment units, coordinate system unification of all data, and normalization of various feature data.

[0023] In this embodiment, obtaining the canopy change mask and change intensity map, generating a canopy gradient response spectrum based on time-series cumulative mapping, and forming a canopy gradient response sequence includes: The remote sensing image data of the same risk assessment unit are acquired in consecutive time phases. Geometric correction, radiometric correction, spatial registration and cropping are performed on the remote sensing images of each time phase, and adjacent time phase image pairs are constructed in chronological order. An improved SNUNet-CD model is constructed, which includes a temporal difference guided coding layer, a gradient enhancement decoding layer, and a response constraint output layer. The temporal difference guided coding layer extracts isomorphic features from adjacent temporal images and generates temporal difference features. The gradient enhancement decoding layer fuses temporal difference features and canopy texture features at different scales step by step. The response constraint output layer outputs the fused features in a dual-branch manner. The SNUNet-CD model is improved by inputting adjacent temporal image pairs. A time-difference guided coding layer is used to extract corresponding layer features from adjacent temporal images, and temporal difference guided feature maps are constructed between each corresponding layer. These temporal difference guided feature maps are then input into a gradient enhancement decoding layer for progressive upsampling and skip connections to obtain a canopy variation representation feature map. Specifically, the construction of temporal difference guided feature maps between each corresponding layer involves: The features output from the previous and subsequent temporal images at the same coding layer are aligned in position. The response differences between the two in the channel dimension are compared pixel by pixel. The feature change amplitude, change direction and local texture differences are jointly encoded to generate temporal difference guiding features to characterize the temporal change degree of the coding layer. Then, based on the temporal difference guiding features, the original features of the corresponding layer are processed to enhance the changed region and suppress the non-changed region to obtain the temporal difference guiding feature map of each corresponding layer. The canopy change feature map is input into the response constraint output layer. One output branch performs pixel-level classification of the changed regions to obtain a canopy change mask, and the other output branch performs hierarchical characterization of the degree of change in the changed regions to obtain a change intensity map. The canopy change mask and change intensity map are then mapped according to risk assessment units to obtain the change range and change intensity results corresponding to each risk assessment unit. The canopy change mask is obtained by inputting the canopy change feature map into the output branch of the change mask, distinguishing between changes and non-changes for each pixel in the canopy change feature map, generating a binarized result representing the spatial distribution of the change region, and using the binarized result as the canopy change mask. The change intensity map is obtained by inputting the canopy change characterization feature map into the change intensity output branch, estimating the degree of change of pixels in the identified change area, generating the corresponding intensity distribution result according to the pixel change response size, and using the intensity distribution result as the change intensity map. Mapping the canopy change mask and change intensity map according to the risk assessment unit: Specifically, the canopy change mask and change intensity map are divided into zones according to the spatial boundaries of the risk assessment unit, and the distribution range and change intensity distribution results of the changed pixels in each risk assessment unit are extracted to form the change range and change intensity results of the corresponding risk assessment unit. By performing time-series cumulative mapping on the canopy change mask and change intensity map corresponding to each adjacent time phase, the change range, change intensity and change location from the previous time phase to the current time phase are superimposed and correlated to generate a canopy gradual change response map reflecting the process of canopy anomaly from initial appearance to continuous expansion. Based on the canopy gradual change response map, the initial appearance time, duration, expansion rate, plaque connectivity growth rate and directional migration characteristics of the canopy anomaly are extracted and correlated and encoded in time sequence to form a canopy gradual change response sequence corresponding to each risk assessment unit.

[0024] In this embodiment, the generation of the pest hysteresis closed-loop index includes: Based on forest environment sensor data, microclimate features related to pest induction were extracted, and a microclimate induction sequence was constructed according to a unified time scale. Based on pest monitoring data, pest population change features and pest source activity features were extracted, and a pest source activity sequence was constructed according to a unified time scale. The canopy gradual response sequence was called as the symptomatic response status of each risk assessment unit in a continuous time series. The microclimate-induced sequences, insect activity sequences, and canopy gradual response sequences were synchronized and aligned. Ternary state units were constructed according to their temporal order, with each ternary state unit including the environmental induced state, insect activity state, and symptomatic response state at the same time. Based on all ternary state units, cross-modal hysteresis maps corresponding to each risk assessment unit were generated. Specifically: Connect each ternary state unit in chronological order, and sequentially link the environmental induced state, insect activity state, and symptomatic response state at adjacent times to form a temporal state chain describing the evolution of the state over time. In the temporal state chain, the relationship between the change from the environmental induced state to the insect source activity state and the relationship between the change from the insect source activity state to the symptomatic response state are used as connection rules to establish directed association relationships between different states, forming a hysteresis association structure that includes multi-stage transmission relationships. The hysteresis-related structures are classified according to risk assessment units. All temporal state chains and their relationships within the same risk assessment unit are integrated to construct a cross-modal hysteresis map that can characterize the temporal transmission and hysteresis relationship between environmental induction, insect activity and symptomatic response. Hysteresis stage encoding was performed on the cross-modal hysteresis map. The state changes in the microclimate-induced sequence were used as the initial drivers, the state changes in the insect activity sequence were used as the intermediate drivers, and the state changes in the canopy gradual response sequence were used as the terminal responses. The phase relationship, closed-loop features, and stage transition features among microclimate induction, insect activity, and canopy gradual response were extracted. The phase relationship was used to characterize the sequential changes from the environmentally induced state to the insect activity state and from the insect activity state to the symptomatic response state. The closed-loop features were used to characterize the head-to-tail connection relationship of the three states in the continuous time series. The stage transition features were used to characterize the transition sequence and duration of the three states from low response to high response. Closed-loop constraint encoding is performed on the cross-modal hysteresis map. Following the order of microclimate induction first, insect source activity followed, and canopy gradual change response manifested later, state chains that do not meet the order are suppressed, while state chains that meet the order and are continuously maintained are enhanced. Based on the enhanced state chains, the hysteresis closed-loop strength, closed-loop duration, and closed-loop stability are extracted to form a hysteresis closed-loop feature set. Based on the hysteresis closed-loop feature set, the phase relationship consistency, closed-loop continuity, and stage transition integrity of each risk assessment unit are jointly encoded to generate the pest hysteresis closed-loop index. Specifically, the joint encoding of the phase relationship consistency, closed-loop continuity, and stage transition integrity of each risk assessment unit is as follows: Consistency matching was performed on the sequential changes of microclimate induction state, insect source activity state and canopy gradual response state in each risk assessment unit. The degree of temporal consistency of environmental induction first appearing, insect source activity later appearing, and symptomatic response finally appearing was statistically analyzed to form phase relationship consistency coding results. The continuity of the closed-loop state chain consisting of environmental induced state, insect source activity state and symptomatic response state in each risk assessment unit is detected to identify whether the closed-loop state chain exists continuously at adjacent time points, and the closed-loop continuity coding result is formed based on the duration of continuous existence and the number of interruptions. The integrity of the transition process of the state chain within each risk assessment unit from the environmental induction stage to the insect source activity stage and then to the symptom response stage is determined. This identifies whether there are missing stages, reversed sequences, or interrupted transitions. Based on the degree of completeness of the stage transition, a stage transition integrity coding result is generated. The phase relationship consistency coding result, the closed-loop continuity coding result, and the stage transition integrity coding result are combined to generate the insect pest lag closed-loop index for the corresponding risk assessment unit. The insect pest lag closed-loop index is used to characterize the degree to which a closed loop of insect pest occurrence is formed between microclimate induction, insect source activity, and canopy gradual change response within each risk assessment unit.

[0025] In this embodiment, the calculation of the pest transmission path and the seepage critical index includes: Using risk assessment units as graph nodes, initial connecting edges are established according to the boundary contact relationship and spatial position relationship between adjacent risk assessment units to form a basic propagation network between risk assessment units, and each connecting edge is assigned a propagation direction identifier. Based on wind direction and speed data, wind direction consistency gates are used to determine the wind direction of each connecting edge in the basic propagation network. The prevailing wind direction is matched with the propagation direction of the connecting edge. The matching connecting edge is given an open state, and the unmatched connecting edge is given a closed state or a suppressed state, forming the first propagation edge set with directional propagation constraints. Based on stand structure data, host continuity gate determination is performed on the first propagation edge set. According to the consistency of host tree species, the continuity of host distribution, and the similarity of stand age among adjacent risk assessment units, connection edges that meet the continuous host condition are assigned an enhanced state, while connection edges that do not meet the continuous host condition are assigned a weakened state, forming a second propagation edge set with host transmission constraints. Specifically, the host continuity gate determination based on stand structure data for the first propagation edge set is as follows: Extract information on host tree species type, host tree species ratio, forest age and canopy closure within adjacent risk assessment units, and identify whether there are the same host tree species and continuously distributed host zones between the two units; Based on the consistency of host tree species, the continuity of host tree species spatial distribution, and the difference in forest age between adjacent risk assessment units, a continuous host adaptation judgment is made for the corresponding connecting edges. Connecting edges with consistent host tree species, continuous host distribution, and forest age difference within a preset range are judged to meet the continuous host condition. The connection edges that meet the continuous host condition are given an enhanced state, the connection edges that do not meet the continuous host condition are given a weakened state, and the processed connection edges are assigned to the second propagation edge set. The forest stand structure data includes tree species type, tree species composition ratio, stand age, canopy closure and spatial distribution information of host tree species, which are used to characterize the forest vegetation structure characteristics and pest host continuity. Based on topographic data, the second propagation edge set is determined by topographic guidance gate. According to the slope aspect connection relationship, slope transition relationship and elevation guidance relationship between adjacent risk assessment units, the connecting edge that is conducive to the spread of pests along the topographic direction is identified. The connecting edge is given a guidance enhancement state, and the connecting edge that is not conducive to the spread is given a blocking state, thus forming a third propagation edge set with topographic guidance constraints. Based on the pest hysteresis closed-loop index, the third propagation edge set is used to determine the phenological synchronization gate. According to whether the hysteresis closed-loop state of adjacent risk assessment units is in an adjacent stage or a continuous stage, the connecting edge with synchronous propagation conditions is identified. The connecting edge that simultaneously satisfies the wind direction consistent gate, host continuous gate, topographic diversion gate and phenological synchronization gate is determined as an effective propagation edge. According to the continuous connection relationship of the effective propagation edge, the switch-type anisotropic seepage propagation map is constructed. Among them, the effective propagation edges that are continuously open in the same time period are combined to form a stage propagation channel. The stage propagation channel that remains open in adjacent time periods is determined as a stable propagation channel. Based on the switch-type anisotropic seepage propagation map, the effective propagation edges from risk assessment units with insect activity indices higher than the preset insect threshold to other risk assessment units are traversed step by step to determine the insect propagation path. The seepage critical index is calculated based on the number of stable propagation channels, the range of risk assessment units covered by the stable propagation channels, and the degree of continuity of the propagation path.

[0026] In this embodiment, the calculation of the initial pest risk value for each risk assessment unit, the inversion of risk-causing factor perturbation for risk assessment units exceeding a preset threshold, the determination of the minimum risk-causing condition set, and the calculation of the risk stability index include: Based on the pest hysteresis closed-loop index, the seepage critical index, and the host susceptibility characteristics of each risk assessment unit, the initial pest risk value of each risk assessment unit is calculated according to the preset risk fusion rules. Risk assessment units whose initial pest risk value reaches the preset risk threshold are selected as units to be inverted. The preset risk fusion rules are based on the degree of influence of the induction stage, the transmission stage, and the host sensitivity stage in the pest occurrence mechanism. Corresponding weights are set for the pest hysteresis closed-loop index, the seepage critical index, and the host susceptibility characteristics, respectively. After the indicators are uniformly scaled, they are weighted and combined. The combination results are subjected to threshold constraint processing to obtain the initial pest risk value of each risk assessment unit. For each unit to be inverted, a set of risk factor disturbances is constructed. The set of risk factor disturbances includes hysteresis closed-loop disturbances, propagation channel disturbances, and host susceptibility disturbances. Hysteresis closed-loop disturbances are used to change the closed-loop continuous state between microclimate induction, insect source activity, and canopy gradual response. Propagation channel disturbances are used to change the continuous state of effective propagation edges in the switch-type anisotropic seepage propagation diagram. Host susceptibility disturbances are used to change the continuous distribution state of the host. The three types of disturbances are generated as candidate disturbance paths in the order of single-factor disturbances, two-factor linkage disturbances, and three-factor linkage disturbances. Each candidate disturbance path is subjected to a step-by-step disturbance inversion of risk factors. The disturbances are applied to the unit to be inverted in order from low disturbance level to high disturbance level. After each level of disturbance, the corresponding risk value is recalculated. The linkage changes of the hysteresis closed loop state, the propagation channel state and the host susceptibility state are recorded during the risk value decrease process, forming a risk factor backtracking chain corresponding to each unit to be inverted. In the risk factor regression chain, candidate disturbance paths that cause the risk value of the unit to be inverted to drop below the preset risk threshold for the first time are screened. The candidate disturbance path with the lowest disturbance level, the fewest number of risk factors involved, and the shortest risk regression path is determined as the minimum risk condition set. The shortest risk regression path means that the number of disturbance levels experienced from the initial pest risk value to below the preset risk threshold is the fewest. The risk stability index is calculated based on the minimum set of risk-causing conditions. The risk stability index is jointly determined by the disturbance level, the number of risk-causing factors, the risk retreat magnitude after the disturbance, and the continuity of the risk-causing factor retreat chain in the minimum set of risk-causing conditions.

[0027] In this embodiment, the output of the corresponding forest pest risk level result and main risk factor information includes: The initial pest risk value, pest lag loop index, seepage critical index and risk stability index of each risk assessment unit are obtained. The indicators are processed to a uniform scale and the indicators of different dimensions are converted into comparable standardized indicator values. Based on the standardized initial pest risk value, pest hysteresis closed-loop index, seepage critical index, and risk stability index, each indicator is weighted and integrated according to preset weights to obtain the comprehensive risk index of each risk assessment unit. The preset weights are as follows: the initial pest risk value has a weight of 0.35, the pest hysteresis closed-loop index has a weight of 0.25, the seepage critical index has a weight of 0.25, and the risk stability index has a weight of 0.15. The sum of the weight coefficients is 1. A risk stability threshold is set, and the comprehensive risk index is corrected based on the comparison between the risk stability index and the risk stability threshold. Specifically, when the risk stability index reaches or exceeds the risk stability threshold, the comprehensive risk index remains unchanged. When the risk stability index is lower than the risk stability threshold, the comprehensive risk index is adjusted downward to obtain the corrected risk index. Multiple risk level classification threshold ranges are set, and the corrected risk index of each risk assessment unit is compared with the threshold range. The corresponding risk level is determined according to the threshold range in which the corrected risk index is located. Based on the risk level of each risk assessment unit and the corresponding pest lag closed-loop index, seepage critical index, and risk stability index, the main causative risk factors are determined, and the risk level results and main causative risk factor information for each risk assessment unit are output. Specifically, the main causative risk factors are determined as follows: Based on the risk level of each risk assessment unit and its corresponding pest lag closed-loop index, seepage critical index and risk stability index, the relative importance of each indicator in the comprehensive risk assessment is analyzed. For each risk assessment unit, by calculating the contribution of each indicator to the final risk level, the indicator with the greatest impact on the change in risk level is identified as the main risk factor. The specific calculation of the contribution of each indicator to the final risk level is as follows: Based on the initial pest risk value, pest lag closed-loop index, seepage critical index and risk stability index of each risk assessment unit, a weighted fusion model is used to calculate the comprehensive risk index and obtain the comprehensive risk score of each unit. By calculating the impact of changes in each indicator in the comprehensive risk score on the final risk level, the partial differential equation method is used to assess the sensitivity of each indicator to changes in the comprehensive risk index, that is, the contribution of each indicator to the final risk level in different scoring intervals. The indicator with the greatest impact is selected as the main risk factor, and the main risk factor and its corresponding value are output as the basis for interpreting the risk assessment results.

[0028] Example 1: To verify the feasibility of this invention in practice, it was applied to a forest farm. The forest farm has a temperate humid climate, and the forest type is mainly mixed coniferous and broad-leaved forest, with the main tree species including Korean pine, spruce, and birch. In recent years, this area has been affected by abnormal climate fluctuations, and the frequency of pine caterpillars and larch caterpillars has increased significantly, especially from June to August, when the pests show a trend of expanding from localized point occurrences to patchy outbreaks. Traditional pest monitoring methods mainly rely on manual patrols and data collection from traps, and risk assessment is conducted through simple weighting or empirical models. This method has a significant lag, and pests can usually only be identified after obvious discoloration or leaf drop in the canopy. Furthermore, it is difficult to accurately determine the direction of pest spread, resulting in a high false alarm rate.

[0029] In this embodiment, the study area was divided into 150 risk assessment units of 80m × 80m each. Thirty sets of forest microclimate sensors and 20 traps were deployed to collect multi-source data. Remote sensing imagery was obtained from Sentinel-2 satellite data, with a temporal resolution of one image every five days, resulting in 24 image acquisitions. The forest sensors collected real-time data on air temperature, relative humidity, soil moisture, and wind speed and direction. Insect population monitoring was conducted using traps to track changes in insect numbers, and the data was preprocessed in conjunction with forest stand structure and topographic data.

[0030] In practical applications, multi-temporal remote sensing images are registered, and an improved SNUNet-CD model is used to detect continuous temporal changes. This model, through time-difference guided coding and a gradient-enhanced decoding structure, enables the continuous capture of subtle canopy changes that were previously difficult to detect, forming a canopy gradient response map that progresses from slight discoloration to local patches to continuous expansion, and further generating a canopy gradient response sequence. Compared to traditional methods that only detect significant changes, this method can identify potential affected areas in advance.

[0031] Cross-modal hysteresis mapping was used to encode the canopy gradual change response sequence, microclimate-induced sequence, and insect activity sequence. In this process, the three types of data were uniformly mapped onto a time-series state chain to identify the sequential relationship between environmental changes, insect population growth, and canopy changes. Experiments revealed that in infestation areas, abnormal temperature and humidity changes typically precede insect population growth by approximately 4 to 6 days, while insect population growth precedes canopy symptoms by approximately 6 to 10 days. This closed-loop relationship can be reliably identified using hysteresis mapping, enabling risk prediction before insect symptoms appear.

[0032] In the spatial analysis phase, a switch-type anisotropic seepage propagation map was constructed based on wind direction, topography, and host distribution. Propagation directions were screened using a consistent wind gate, suitable habitats were identified using a host continuity gate, valley-guided paths were identified using a topographical guidance gate, and effective propagation links were selected by combining phenological synchronization gates. The results showed a significantly improved correlation between pest propagation paths and actual patrol records, enabling accurate prediction of the pest's spread from southwest to northeast.

[0033] In the risk correction process, a risk factor perturbation inversion mechanism is introduced. Units initially identified as high-risk undergo stepwise perturbation analysis. By altering the intensity of insect activity, the connectivity of transmission pathways, and host continuity, the minimum set of risk-causing conditions that reduce risk is identified. Experimental results show that some high-risk units caused by short-term climate anomalies experienced a significant risk reduction after perturbation and were effectively eliminated, reducing false alarms.

[0034] Table 1. Comparison of the effectiveness of forest pest risk assessment methods First identification time of pests (days before manual confirmation) 2 days in advance 8 days in advance Canopy anomaly detection accuracy 76.8% 91.2% Pest risk identification accuracy 79.5% 92.6% High risk false alarm rate 23.4% 9.1% underreporting rate 17.2% 6.8% Pest transmission path matching rate 64.7% 89.5% Success rate of hysteresis relationship identification 41.3% 85.7% Consistency between risk level and actual measurement 78.9% 93.4% Average risk stability 0.57 0.83 Accurate prediction of high-risk area proportion 68.2% 90.6% As can be seen from Table 1, the present invention has significant advantages in early pest identification. The time for the first pest identification is improved from 2 days to 8 days earlier than that of the traditional method. This indicates that by extracting the canopy gradual response and modeling the hysteresis relationship, potential change signals can be captured before pest symptoms appear. The accuracy of canopy anomaly detection is improved from 76.8% to 91.2%, indicating that the improved SNUNet-CD model has a stronger ability to identify subtle changes in multiple time phases and can effectively improve the accuracy of early identification.

[0035] Regarding the accuracy of risk assessment, the method of this invention improves the accuracy of pest risk identification from 79.5% to 92.6%, while reducing the false alarm rate of high risk from 23.4% to 9.1% and the false negative rate from 17.2% to 6.8%. This indicates that the perturbation inversion mechanism of risk-causing factors can effectively eliminate interference from non-pest factors, improve the reliability of assessment results, and achieve a 93.4% consistency between the risk level and the actual measurement, further verifying the high consistency between the assessment results and the actual situation.

[0036] In terms of spatial propagation and mechanism characterization capabilities, the method of this invention increases the pest propagation path matching rate from 64.7% to 89.5%, the success rate of hysteresis relationship identification from 41.3% to 85.7%, the average risk stability from 0.57 to 0.83, and the proportion of accurate area prediction for high-risk areas from 68.2% to 90.6%. This demonstrates that the invention can accurately characterize the evolution of pests in time and space, and improve the stability and spatial prediction capabilities of risk assessment.

[0037] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A forest pest risk grade intelligent assessment method based on multi-source sensing data, characterized in that, include: Acquire multi-source sensor data of the target forest area, preprocess the multi-source sensor data, and obtain standardized datasets corresponding to each risk assessment unit; For remote sensing image data acquired in consecutive time phases for the same risk assessment unit, an improved SNUNet-CD model is used to detect changes in adjacent time phases, resulting in a canopy change mask and change intensity map. Based on the time-series cumulative mapping, a canopy gradual response map is generated, forming a canopy gradual response sequence. Based on standardized datasets, microclimate-induced sequences and insect activity sequences were constructed. Combined with canopy gradual response sequences, cross-modal hysteresis map encoding was performed to extract the phase shift relationship, closed-loop evolution trajectory and stage transition characteristics between microclimate-induced, insect activity and canopy gradual response, and generate insect hysteresis closed-loop index. Based on the spatial propagation influence factors between risk assessment units, an on / off anisotropic seepage propagation map is constructed, which includes wind direction consistent gate, host continuity gate, topographic diversion gate and phenological synchronous gate, and the pest propagation path and seepage critical index are calculated. Based on the pest lag closed-loop index, seepage critical index and host susceptibility characteristics, the initial pest risk value of each risk assessment unit is calculated. For risk assessment units with risk values ​​higher than the preset threshold, the perturbation inversion of risk-causing factors is performed to determine the minimum set of risk-causing conditions and calculate the risk stability index. Based on the initial pest risk value, pest lag closed-loop index, seepage critical index, and risk stability index, the risk level of each risk assessment unit is classified, and the corresponding forest pest risk level results and information on the main risk factors are output.

2. The intelligent evaluation method of forest pest risk level based on multi-source sensing data according to claim 1, characterized in that, The multi-source sensor data includes remote sensing image data, forest environment sensor data, insect monitoring data, meteorological data, forest stand structure data, and topographic data. 3.The intelligent forest pest risk rating method based on multi-source sensing data according to claim 1, wherein, The preprocessing of multi-source sensor data includes outlier removal and missing value completion for various types of data, time alignment of data from different sources according to a unified time scale, spatial mapping of data with different spatial resolutions according to risk assessment units, coordinate system unification of all data, and normalization of various feature data. 4.The intelligent forest pest risk rating method based on multi-source sensing data according to claim 1, wherein, The obtained canopy change mask and change intensity map are used to generate a canopy gradient response spectrum based on time-series cumulative mapping, forming a canopy gradient response sequence, including: The remote sensing image data of the same risk assessment unit are acquired in consecutive time phases. Geometric correction, radiometric correction, spatial registration and cropping are performed on the remote sensing images of each time phase, and adjacent time phase image pairs are constructed in chronological order. An improved SNUNet-CD model is constructed, which includes a temporal difference guided coding layer, a gradient enhancement decoding layer, and a response constraint output layer. The temporal difference guided coding layer extracts isomorphic features from adjacent temporal images and generates temporal difference features. The gradient enhancement decoding layer fuses temporal difference features and canopy texture features at different scales step by step. The response constraint output layer outputs the fused features in a dual-branch manner. The SNUNet-CD model is improved by inputting adjacent temporal image pairs. The corresponding layer features of adjacent temporal images are extracted using the time difference guided coding layer. Temporal difference guided feature maps are constructed between each corresponding layer. The temporal difference guided feature maps of each layer are then input into the gradient enhancement decoding layer for step-by-step upsampling and skip connection fusion to obtain the canopy change characterization feature map. The canopy change feature map is input into the response constraint output layer. One output branch performs pixel-level classification of the change area to obtain the canopy change mask, and the other output branch performs hierarchical characterization of the degree of change in the change area to obtain the change intensity map. The canopy change mask and the change intensity map are mapped according to the risk assessment unit to obtain the change range and change intensity results corresponding to each risk assessment unit. By performing time-series cumulative mapping on the canopy change mask and change intensity map corresponding to each adjacent time phase, the change range, change intensity and change location from the previous time phase to the current time phase are superimposed and correlated to generate a canopy gradual change response map reflecting the process of canopy anomaly from initial appearance to continuous expansion. Based on the canopy gradual change response map, the initial appearance time, duration, expansion rate, plaque connectivity growth rate and directional migration characteristics of the canopy anomaly are extracted and correlated and encoded in time sequence to form a canopy gradual change response sequence corresponding to each risk assessment unit.

5. The intelligent assessment method of forest pest risk level based on multi-source sensing data according to claim 1, characterized in that, The generated pest lag closed-loop index includes: Based on forest environment sensor data, microclimate features related to pest induction were extracted, and a microclimate induction sequence was constructed according to a unified time scale. Based on pest monitoring data, pest population change features and pest source activity features were extracted, and a pest source activity sequence was constructed according to a unified time scale. The canopy gradual response sequence was called as the symptomatic response status of each risk assessment unit in a continuous time series. The microclimate-induced sequence, insect activity sequence, and canopy gradual response sequence were synchronized and aligned. Ternary state units were constructed according to their temporal order. Each ternary state unit includes the environmental induced state, insect activity state, and symptomatic response state at the same time. Based on all ternary state units, cross-modal hysteresis maps corresponding to each risk assessment unit were generated. Hysteresis stage encoding was performed on the cross-modal hysteresis map. The state changes in the microclimate-induced sequence were used as the initial driver, the state changes in the insect source activity sequence were used as the relay, and the state changes in the canopy gradual response sequence were used as the terminal response. The phase relationship, closed-loop characteristics and stage transition characteristics between microclimate induction, insect source activity and canopy gradual response were extracted. Closed-loop constraint encoding is performed on the cross-modal hysteresis map. Following the order of microclimate induction first, insect source activity followed, and canopy gradual change response manifested later, state chains that do not meet the order are suppressed, while state chains that meet the order and are continuously maintained are enhanced. Based on the enhanced state chains, the hysteresis closed-loop strength, closed-loop duration, and closed-loop stability are extracted to form a hysteresis closed-loop feature set. Based on the hysteresis closed-loop feature set, the phase relationship consistency, closed-loop continuity and stage transition integrity of each risk assessment unit are jointly encoded to generate the pest hysteresis closed-loop index.

6. The intelligent assessment method for forest pest risk level based on multi-source sensor data according to claim 1, characterized in that, The calculation of pest transmission pathways and seepage critical index includes: Using risk assessment units as graph nodes, initial connecting edges are established according to the boundary contact relationship and spatial position relationship between adjacent risk assessment units to form a basic propagation network between risk assessment units, and each connecting edge is assigned a propagation direction identifier. Based on wind direction and speed data, wind direction consistency gates are used to determine the wind direction of each connecting edge in the basic propagation network. The prevailing wind direction is matched with the propagation direction of the connecting edge. The matching connecting edge is given an open state, and the unmatched connecting edge is given a closed state or a suppressed state, forming the first propagation edge set with directional propagation constraints. Based on stand structure data, host continuity gate determination is performed on the first propagation edge set. According to the consistency of host tree species, the continuity of host distribution, and the similarity of stand age between adjacent risk assessment units, the connection edge that meets the continuous host condition is given an enhanced state, and the connection edge that does not meet the continuous host condition is given a weakened state, thus forming a second propagation edge set with host transmission constraints. Based on topographic data, the second propagation edge set is determined by topographic guidance gate. According to the slope aspect connection relationship, slope transition relationship and elevation guidance relationship between adjacent risk assessment units, the connecting edge that is conducive to the spread of pests along the topographic direction is identified. The connecting edge is given a guidance enhancement state, and the connecting edge that is not conducive to the spread is given a blocking state, thus forming a third propagation edge set with topographic guidance constraints. Based on the pest hysteresis closed-loop index, the third propagation edge set is used to determine the phenological synchronization gate. According to whether the hysteresis closed-loop state of adjacent risk assessment units is in an adjacent stage or a continuous stage, the connecting edge with synchronous propagation conditions is identified. The connecting edge that simultaneously satisfies the wind direction consistent gate, host continuous gate, topographic diversion gate and phenological synchronization gate is determined as an effective propagation edge. According to the continuous connection relationship of the effective propagation edge, the switch-type anisotropic seepage propagation map is constructed. Among them, the effective propagation edges that are continuously open in the same time period are combined to form a stage propagation channel. The stage propagation channel that remains open in adjacent time periods is determined as a stable propagation channel. Based on the switch-type anisotropic seepage propagation map, the effective propagation edges from risk assessment units with insect activity indices higher than the preset insect threshold to other risk assessment units are traversed step by step to determine the insect propagation path. The seepage critical index is calculated based on the number of stable propagation channels, the range of risk assessment units covered by the stable propagation channels, and the degree of continuity of the propagation path.

7. The intelligent assessment method of forest pest risk level based on multi-source sensing data according to claim 1, characterized in that, The calculation of the initial pest risk value for each risk assessment unit, the inversion of risk-causing factor perturbation for risk assessment units exceeding a preset threshold, the determination of the minimum risk-causing condition set, and the calculation of the risk stability index include: Based on the pest lag closed-loop index, seepage critical index, and host susceptibility characteristics of each risk assessment unit, the initial pest risk value of each risk assessment unit is calculated according to the preset risk fusion rules, and the risk assessment units whose initial pest risk value reaches the preset risk threshold are selected as the inversion units. For each unit to be inverted, a set of risk-causing disturbances is constructed. The set of risk-causing disturbances includes hysteresis closed-loop disturbances, propagation channel disturbances, and host susceptibility disturbances. Candidate disturbance paths are generated by taking the three types of disturbances in the order of single-factor disturbances, two-factor linkage disturbances, and three-factor linkage disturbances. Each candidate disturbance path is subjected to a step-by-step disturbance inversion of risk factors. The disturbances are applied to the unit to be inverted in order from low disturbance level to high disturbance level. After each level of disturbance, the corresponding risk value is recalculated. The linkage changes of the hysteresis closed loop state, the propagation channel state and the host susceptibility state are recorded during the risk value decrease process, forming a risk factor backtracking chain corresponding to each unit to be inverted. In the risk factor backtracking chain, candidate disturbance paths that cause the risk value of the unit to be inverted to drop below the preset risk threshold for the first time are selected, and the candidate disturbance path with the lowest disturbance level, the fewest number of risk factors involved and the shortest risk backtracking path is determined as the minimum risk condition set. The risk stability index is calculated based on the minimum set of risk-causing conditions. The risk stability index is jointly determined by the disturbance level, the number of risk-causing factors, the risk retreat magnitude after the disturbance, and the continuity of the risk-causing factor retreat chain in the minimum set of risk-causing conditions.

8. The intelligent assessment method for forest pest risk level based on multi-source sensor data according to claim 1, characterized in that, The output includes the corresponding forest pest risk level results and main risk factor information, including: The initial pest risk value, pest lag loop index, seepage critical index and risk stability index of each risk assessment unit are obtained. The indicators are processed to a uniform scale and the indicators of different dimensions are converted into comparable standardized indicator values. Based on the standardized initial pest risk value, pest lag closed-loop index, seepage critical index and risk stability index, the indicators are weighted and integrated according to the preset weights to obtain the comprehensive risk index of each risk assessment unit. A risk stability threshold is set, and the comprehensive risk index is corrected based on the comparison between the risk stability index and the risk stability threshold. Specifically, when the risk stability index reaches or exceeds the risk stability threshold, the comprehensive risk index remains unchanged. When the risk stability index is lower than the risk stability threshold, the comprehensive risk index is adjusted downward to obtain the corrected risk index. Multiple risk level classification threshold ranges are set, and the corrected risk index of each risk assessment unit is compared with the threshold range. The corresponding risk level is determined according to the threshold range in which the corrected risk index is located. Based on the risk level of each risk assessment unit and the corresponding pest lag closed-loop index, seepage critical index and risk stability index, the main risk factors are determined, and the risk level results and main risk factor information of each risk assessment unit are output.

Images