A method for detecting and analyzing the salt leaching effect of a coastal heavy saline soil salt barrier layer arrangement

Abstract

The present application relates to the technical field of saline soil improvement and monitoring, and specifically discloses a method for detecting and analyzing the effect of salt leaching of a coastal severe saline soil salt-blocking layer, which comprises: an intelligent sensing system deployed based on a research area, a distributed soil conductivity sensing array in the intelligent sensing system being used to obtain the spatial distribution of soil conductivity above the salt-blocking layer, an initial background field of salt leaching effect being constructed, and a potential failure area being located; an improved normalized difference water index and salt index spatial fusion algorithm being combined, image features of preferential flow paths being located by image recognition, and a failure area of the salt-blocking layer, i.e., a locally damaged or unevenly laid area, being determined; and high-density conductivity data above the salt-blocking layer being obtained by the distributed soil conductivity sensing array, an initial background field of salt leaching effect being constructed in combination with spatial interpolation and abnormal threshold extraction, and an abnormally high conductivity area being accurately identified, so that the defect that a traditional overall sampling cannot locate a local failure area is overcome, and a clear spatial direction is provided for repair.

Description

Technical Field

[0001] This invention relates to the field of saline soil improvement and monitoring technology, and in particular to a method for detecting and analyzing the salt leaching effect of salt isolation layer deployment in severely saline coastal soil. Background Technology

[0002] Severely saline coastal soil is a typical type of salinized soil, typically characterized by high salt content, which seriously affects the physical and chemical properties of the soil and the plant growth environment. In severely saline coastal soil, salt accumulation may lead to reduced soil permeability, nutrient loss, and inhibition of plant roots, thereby affecting crop planting and growth. To improve the utilization and quality of saline soil, salt barrier technology is used to prevent salt from migrating to deeper soil layers, which is an effective soil improvement method. Salt leaching effect detection and analysis is a means of evaluating the effectiveness of this improvement measure, which can scientifically determine the distribution changes of salt in the soil and the inhibitory effect of the salt barrier on salt migration.

[0003] In existing technologies, it is difficult to accurately locate local failure areas in the initial stage of salt leaching after the salt barrier layer is laid. Current detection methods rely heavily on overall sampling, which makes it difficult to quickly identify preferential flow channels formed by local damage or uneven laying of the salt barrier layer. This results in a lack of spatial orientation in the evaluation of the improvement effect. Moreover, traditional static sampling cannot distinguish between capillary action and tidal backwater as the dominant mechanisms for salt return, leading to unclear dynamics of the salt return process and a lack of targeted subsequent treatment measures. Furthermore, due to the lack of decision-making basis for the timing and method of remediation based on the classification of salt return mechanisms, existing remediation strategies often fail to achieve accurate restoration of the salt barrier layer function due to inappropriate timing or mismatch of methods. Therefore, this paper proposes a method for detecting and analyzing the salt leaching effect of salt barrier layer deployment in severely saline coastal soils to solve the above-mentioned problems. Summary of the Invention

[0004] To overcome the shortcomings of the prior art, the present invention provides a method for detecting and analyzing the salt leaching effect of salt isolation layer layout in heavily saline coastal soil, which can effectively solve the problems involved in the prior art.

[0005] The objective of this invention can be achieved through the following technical solution: This invention provides a method for detecting and analyzing the salt leaching effect of salt-isolating layers in heavily saline coastal soil, comprising the following steps: Step 1: Based on the intelligent sensing system deployed in the study area, the distributed soil conductivity sensing array is used to obtain the spatial distribution of soil conductivity above the salt barrier layer, construct the initial background field of salt leaching effect, locate potential failure areas, realize rapid initial screening of failure areas, and reduce the scope of blind detection. Step 2: Combining the improved spatial fusion algorithm of normalized differential water index and salinity index, the image features of the priority flow path are located by image recognition to determine the area of ​​salt barrier failure, that is, the area of ​​local damage or uneven laying, accurately identify the priority flow channel and improve the spatial resolution of failure location. Step 3: Deploy layered temperature, humidity and salinity sensors in the identified failure areas, and simultaneously access real-time tidal water level data and groundwater depth monitoring information to establish a dynamic monitoring network for the salt return process, construct a multi-source dynamic sensing system, and support the refined analysis of the salt return process. Step 4: Use time-series clustering algorithm to separate capillary action and tidal backflow driven salt return modes, analyze the dominant salt return mechanism in different regions, clarify the failure cause type, quantitatively distinguish the salt return driving mechanism, and avoid blindly applying treatment measures. Step 5: Based on the classification results of the salt return dominant mechanism, construct a soil salinity change prediction model, integrate soil salinity dynamics, meteorological and tidal cycle data, output the remediation window period, realize quantitative prediction of remediation timing, and avoid unfavorable windows of tidal evaporation. Step 6: Combining the output repair window period, use reinforcement learning algorithm to optimize and output the optimal salt barrier layer replacement scheme, realize the accurate repair and functional restoration of the salt barrier layer failure area, form a closed-loop detection-decision-control system, and continuously improve the accuracy of repair decision-making.

[0006] Preferably, step 1 specifically includes: A smart sensing system was deployed in a 5m×5m grid within the study area. The distributed soil conductivity sensor array included in the smart sensing system was used to collect apparent conductivity data of the soil layer 0-30cm above the salt barrier layer, obtain spatially continuous conductivity distribution information, and ensure that the conductivity data is spatially continuous and covers the entire monitoring area. The ordinary Kriging interpolation method is used to spatially reconstruct the collected soil apparent electrical conductivity data to generate an initial background field for salt leaching effect. The mean and standard deviation of the background field are calculated to identify areas with abnormally high electrical conductivity values, thereby visualizing the spatial distribution of salt leaching effect and effectively highlighting areas with abnormal salt content. Using the background mean plus 1.5 times the standard deviation as a threshold, continuous spatial units above the threshold are extracted and preliminarily identified as potential failure areas of the salt barrier layer. This information is used to guide the subsequent identification of priority flow channels, quickly locate areas with insufficient salt leaching effect, and provide spatial targeting for subsequent detection.

[0007] Preferably, step 2 specifically includes: High-resolution multispectral remote sensing images of the study area were acquired, the improved normalized differential water index was calculated, and the spatial distribution of the surface wet zone was extracted as a water indicator feature of the priority flow path, effectively identifying the surface wet area formed by the seepage of the salt barrier layer. Simultaneously calculate the salinity index, extract the spatial distribution of soil salinity enrichment areas, and spatially overlay it with the distribution of the humid zone to identify abnormal salt-moisture coupling patches and accurately locate abnormal areas where water and salt accumulate simultaneously. A semantic segmentation network is used to perform image recognition on superimposed salt-wet coupling abnormal patches, extract the image features of the priority flow path with strip or network distribution, determine the local damage or uneven laying area of ​​the salt barrier layer, automatically identify the geometric shape of the priority flow channel, and determine the type of failure area.

[0008] Preferably, step 2 further includes: The improved normalized differential water index and salinity index are normalized at the pixel level to eliminate dimensional differences and generate a dual-channel fused feature map, thereby eliminating dimensional differences and improving the accuracy and stability of subsequent feature extraction. Principal component analysis is used to reduce the dimensionality and enhance the features of the dual-channel fusion feature map. The first principal component is extracted as the salt-wet coupling strength index to highlight the spatial signal of the preferred flow channel, effectively enhance the preferred flow channel signal, and suppress background noise interference. The salt-wet coupling strength index is input into the trained random forest classifier to automatically segment the priority flow path region and output the spatial vector boundary of the salt barrier failure region, thus realizing automatic segmentation of the failure region.

[0009] Preferably, step 3 specifically includes: Layered temperature, humidity and salt sensors are deployed in the identified salt barrier failure area, with monitoring depths of at least 5cm, 15cm, 25cm and 40cm below the salt barrier layer, to obtain vertical profile data of soil temperature, humidity and electrical conductivity, so as to achieve accurate tracking and stratified quantification of the vertical migration path of salt. By simultaneously accessing real-time tidal water level data of the study area and the burial depth and salinity information of nearby groundwater monitoring wells, a multi-source data synchronous acquisition link of surface-groundwater-tidal is formed, a multi-factor collaborative observation system for the salt return process is constructed, and the mechanism analysis capability is improved. Set the sensor data acquisition frequency to no less than once per hour, and build a dynamic monitoring network for the salt return process covering the entire tidal cycle. This network will support the time-series analysis of the salt return mode, ensure continuous capture of the salt return dynamics throughout the entire cycle, and eliminate monitoring blind spots.

[0010] Preferably, step 4 specifically includes: Extract time-series data of soil electrical conductivity at various depths for at least 15 consecutive days from the dynamic monitoring network of the salt return process, construct a multidimensional time series sample set, and each sample contains depth and time dimension information to achieve complete preservation of the vertical dynamic characteristics of the salt return process; The dynamic time warping algorithm is used to calculate the similarity distance between time series samples at different locations, eliminate the impact of phase shift caused by tides and evaporation on clustering, eliminate phase interference caused by tides and evaporation, and improve clustering accuracy. By combining the K-means clustering algorithm, time-series samples are divided into two types of salt return dominant mechanisms: capillary action-dominated and tidal support-dominated. Salt return type labels for each failure region are output, thereby realizing the automatic identification and classification of salt return dominant mechanisms.

[0011] Preferably, step 4 further includes: For areas identified as capillary-dominated, the temporal correlation between soil salinity changes and evaporation and groundwater depth was analyzed to identify the gradual salt return characteristics induced by uneven salt barrier layer laying, accurately distinguish the causes of gradual salt return, and avoid confusion with damage-type salt return. For areas identified as being dominated by tidal backwater, the phase difference and time lag coefficient between the peak salinity and the high tide event are calculated to identify the pulsed salt return characteristics induced by local damage to the salt barrier layer, quickly locate the pulsed salt return caused by the damage, and provide a basis for targeted repair. Based on the combined salt return pattern type and spatial distribution, the failure causes are classified into uneven laying type, damage type, or tidal induced type, which serves as the basis for subsequent repair strategy selection, realizing a closed loop of failure cause classification and improving the matching degree between repair strategies and on-site working conditions.

[0012] Preferably, step 5 specifically includes: Based on the classification results of the salt return dominant mechanism, a multi-dimensional feature input set integrating soil salinity dynamics, daily evaporation, precipitation and tidal cycle data is constructed to form a remediation decision sample library, establish high-quality training samples, and improve the model's prediction reliability and generalization ability. A soil salinity change prediction model was constructed using a long short-term memory network. The model learned the nonlinear mapping relationship between the salt return rate and meteorological-tidal factors, and predicted the soil salinity change trend in the next 7 to 15 days, so as to realize the early prediction of the salt return trend and gain sufficient preparation time for remediation operations. Based on the condition that the predicted salinity return is below the threshold and lasts for at least 72 hours, the optimal repair window is output. At the same time, the high tide and high evaporation periods that are not suitable for repair are marked, so as to accurately avoid unfavorable periods and significantly improve the success rate of salt barrier repair operations.

[0013] Preferably, step 6 specifically includes: The spatial distribution, failure cause type, and repair window period of the salt barrier failure area are used as input to the state space. The location, depth, thickness, and material type of the replacement are defined as the action space to realize the quantitative characterization and decision modeling of the multi-attribute state of the failure area. A multi-objective reward function is constructed with the goal of minimizing the salt return rate after repair and maximizing the salt isolation function recovery period. The Q-learning algorithm is used to iteratively train the repair strategy to balance salt suppression and functional durability and improve the overall performance of the strategy. Output the optimal salt barrier layer replacement plan, including the spatial vector of the replacement area, layer structure parameters, and drainage pipe adjustment suggestions, to guide precise on-site repair construction, generate a refined repair plan that can be directly constructed, and improve on-site operation efficiency.

[0014] Preferably, step 6 further includes: After the salt barrier layer is repaired, the distributed soil conductivity sensor array is reactivated to collect spatial distribution data of soil conductivity above the repaired salt barrier layer, so as to achieve rapid and accurate verification of salt distribution after repair. The difference analysis was performed between the repaired data and the initial background field to calculate the salt reduction rate and the degree of improvement in spatial uniformity of each failure area, evaluate the repair effect, quantitatively determine the repair level and identify residual failure risks. The results of the remediation effect assessment are fed back to the soil salinity change prediction model to update the sample library and model parameters, realizing closed-loop iteration and adaptive optimization of the entire process of detection, decision-making and regulation, and continuously improving the model prediction accuracy and the adaptability of the remediation strategy.

[0015] Compared with the prior art, the beneficial effects of the present invention are: 1. This method for detecting and analyzing the salt leaching effect of salt barrier layer deployment in severely saline coastal soil uses a distributed soil conductivity sensor array to acquire high-density conductivity data above the salt barrier layer. Combined with spatial interpolation and anomaly threshold extraction, an initial background field for salt leaching effect is constructed. This method can accurately identify areas with abnormally high conductivity values, overcome the shortcomings of traditional whole-sampling that cannot locate local failure areas, and provide clear spatial guidance for remediation.

[0016] 2. This method for detecting and analyzing the salt leaching effect of salt barrier layer deployment in severely saline coastal soil integrates an improved normalized difference water index and salinity index. By extracting strip-shaped or mesh-shaped preferential flow paths through image recognition and semantic segmentation networks, and combining principal component analysis and random forest classification, it can automatically determine areas of local damage or uneven deployment of the salt barrier layer, significantly improving the accuracy and automation level of salt barrier layer failure cause identification.

[0017] 3. This method for detecting and analyzing the salt leaching effect of salt isolation layer in heavily saline coastal soil constructs a dynamic monitoring network for the salt return process through layered temperature, humidity and salt sensors. It uses a time-series clustering algorithm to separate the salt return modes driven by capillary action and tidal backing. Combined with hysteresis correlation analysis and time delay coefficient calculation, it can accurately identify the characteristics of gradual salt return and pulsed salt return, and achieve a refined classification of failure causes.

[0018] 4. This method for detecting and analyzing the salt leaching effect of salt isolation layer deployment in severely saline coastal soils constructs a long short-term memory network prediction model based on the classification results of salt return mechanism. It integrates soil salinity dynamics, meteorological and tidal cycle data to predict future soil salinity change trends. Based on the salt return threshold and precipitation and tidal conditions, it intelligently outputs the best repair window period and the period when repair is not suitable to guide the construction schedule. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the workflow for detecting and analyzing the salt leaching effect of a salt barrier layer in heavily saline coastal soil according to the present invention. Figure 2 This is a schematic flowchart of a method for detecting and analyzing the salt leaching effect of a salt barrier layer in heavily saline coastal soil according to the present invention. Detailed Implementation

[0020] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are some embodiments of the present invention, but not all embodiments.

[0021] Example 1, please refer to Figure 1 , Figure 2 This invention provides a technical solution: a method for detecting and analyzing the salt leaching effect of salt-isolating layers in heavily saline coastal soil, comprising the following steps: Step 1: Based on the intelligent sensing system deployed in the study area, the distributed soil conductivity sensor array is used to acquire the spatial distribution of soil conductivity above the salt barrier layer, constructing an initial background field for salt leaching effect, locating potential failure zones, and achieving rapid initial screening of failure areas to reduce the scope of blind detection. The intelligent sensing system is deployed in a 5m×5m grid within the study area. Utilizing the distributed soil conductivity sensor array within the intelligent sensing system, apparent conductivity data of the 0–30cm soil layer above the salt barrier layer are collected to obtain spatially continuous conductivity distribution information, ensuring spatial continuity of conductivity data. Continuing, covering the entire monitoring area, the collected apparent electrical conductivity data of the soil layer were spatially reconstructed using the ordinary Kriging interpolation method to generate an initial background field for salt leaching effect. The mean and standard deviation of the background field were calculated to identify areas with abnormally high electrical conductivity values, thus visualizing the spatial distribution of salt leaching effect and effectively highlighting areas with abnormal salt content. Using the mean of the background field plus 1.5 times the standard deviation as a threshold, continuous spatial units above the threshold were extracted and preliminarily identified as potential failure areas of the salt barrier layer. This information was used to guide the subsequent identification of priority flow channels, quickly pinpoint areas with insufficient salt leaching effect, and provide spatial targeting for subsequent detection. It should be noted that within the study area, a distributed soil conductivity sensor array was deployed with a grid spacing of 5m × 5m. Each grid node was equipped with a four-ring conductivity probe, with the probe measurement depth set to 0–30cm, corresponding to the cultivated layer above the salt barrier. Before deployment, each probe was uniformly calibrated using a standard potassium chloride solution to ensure measurement errors were controlled within ±2%. Data collection was conducted under conditions of three consecutive days without rainfall or artificial irrigation to eliminate interference from moisture disturbance on conductivity readings. Data was collected three times consecutively at each measuring point, and the results were calculated. The technical mean was used as the apparent conductivity value at that point. The coordinates of all measurement points were determined using RTK-GPS positioning, with a planar positioning accuracy better than 5 cm to ensure spatial accuracy. After collecting data from all grid measurement points, a spatial discrete conductivity dataset covering the study area was formed. Ordinary Kriging interpolation was used to spatially reconstruct the discrete conductivity dataset. A spherical model was selected for the variogram model, with the nugget value set to 0. The sill value was determined based on the experimental variogram fitting. The search radius was set to 1.5 times the spacing between adjacent grids, i.e., 7.5 m. After interpolation, the pixel size was generated. A continuous spatial grid of 0.5m × 0.5m was used as the initial background field for the salt leaching effect. The conductivity values ​​of all pixels within the background field were statistically analyzed, and their arithmetic mean μ and standard deviation σ were calculated. An anomaly threshold of μ + 1.5σ was set. Each pixel in the background field was traversed, and the set of pixels with conductivity greater than this threshold was extracted. Connectivity analysis was performed on the extracted pixels, and isolated noise patches with an area less than 1m² were deleted. Continuous spatial cells were retained as areas with abnormally high conductivity values. These retained areas with abnormally high conductivity values ​​were preliminarily identified as potential failure zones of the salt barrier layer, based on the conductivity of the region. Significantly higher than the average level of the background field indicates insufficient salt leaching or preferential salt accumulation. Potential failure areas are output in the form of vector polygons, with each polygon accompanied by attribute information, including area, average conductivity, maximum conductivity, and anomaly intensity index. The anomaly intensity index is defined as the ratio of the average conductivity within the area to the average value of the background field. The scale of the output potential failure area spatial distribution map is no less than 1:1000, which is used to guide the cropping of the priority flow channel identification range of the remote sensing image in subsequent steps, ensuring that high-resolution detection resources are concentrated in key areas and reducing unnecessary computation. Step 2: Combining the improved Normalized Differential Water Index (NDI) and Salinity Index spatial fusion algorithm, the algorithm locates the image features of the priority flow path through image recognition, determines the failure area of ​​the salt barrier layer (i.e., areas of local damage or uneven laying), accurately identifies the priority flow channel, improves the spatial resolution of failure location, acquires high-resolution multispectral remote sensing images of the study area, calculates the improved NDI, extracts the spatial distribution of the surface humid zone as a moisture indicator feature of the priority flow path, effectively identifies the surface humid area formed by salt barrier layer leakage, simultaneously calculates the salinity index, extracts the spatial distribution of soil salt enrichment area, and spatially overlays it with the humid zone distribution to identify salt-moisture coupling anomalous patches, accurately locates anomalous areas where water and salt accumulate simultaneously, uses a semantic segmentation network to perform image recognition on the overlaid salt-moisture coupling anomalous patches, extracts the image features of the priority flow path with strip or mesh distribution, determines the areas of local damage or uneven laying of the salt barrier layer, automatically identifies the geometric shape of the priority flow channel, and determines the type of failure area. It should be noted that after acquiring high-resolution multispectral remote sensing images of the study area, radiometric calibration and atmospheric correction were performed on the images to eliminate the effects of atmospheric scattering and absorption, and the improved normalized difference water index was calculated. The formula is ,in It is in the green light band (center wavelength approximately 560nm). The mid-infrared band (center wavelength approximately 1610 nm) was used to generate an MNDWI raster map, with a value range of -1 to 1. A threshold of 0.2 was set, and pixels above this threshold were extracted as the spatial distribution of the surface humid zone. This distribution serves as a moisture indicator feature of the preferred flow path, used to identify humid areas formed by seepage above the salt barrier layer; the salinity index was calculated simultaneously. The formula is ,in It is in the red light band (center wavelength approximately 665nm). The near-infrared band (center wavelength approximately 865nm) was used to calculate and generate an NDSI raster map. A threshold of 0.3 was set, and pixels exceeding this threshold were extracted as spatial distributions of soil salinity enrichment areas. Spatial overlay analysis was performed on the distributions of the wetting zone and salinity enrichment areas. Pixel-level logical AND operations were used to extract sets of pixels that simultaneously satisfied both MNDWI and NDSI values ​​of 0.3, identifying these as salt-moisture coupling anomalous patches. These patches indicate abnormal regions where both water accumulation and surface salt accumulation exist above the salt barrier layer. DeepLab was used for analysis. The V3+ semantic segmentation network performs image recognition on superimposed salt-wet coupling anomalous patches. The input is a fused feature map containing both MNDWI and NDSI channels, and the output is a pixel-level classification result. The network uses Xception as the backbone feature extraction network, with an output stride of 16 and dilated convolution sampling rates of 6, 12, and 18. The training sample set contains 500 labeled images, labeled as priority flow paths and non-priority flow paths. The priority flow paths are represented by strip-like or mesh-like geometric shapes. The model achieves an average intersection-union ratio of 0.85 on the validation set. When applied, the salt-wet coupling anomalous patch image is input into the model, and the continuous pixel set predicted as a priority flow path is extracted. After morphological closing operations to eliminate small fractures, the output is a strip-like or mesh-like geometric vector, which is determined to be a localized area of ​​damage or uneven distribution of the salt barrier layer. Furthermore, step 2 also includes: performing pixel-level normalization on the improved normalized differential water index and salinity index to eliminate dimensional differences, generating a dual-channel fusion feature map, eliminating dimensional differences, improving the accuracy and stability of subsequent feature extraction, using principal component analysis to reduce the dimensionality and enhance the features of the dual-channel fusion feature map, extracting the first principal component as the salt-wet coupling strength index, highlighting the spatial signal of the priority flow channel, effectively enhancing the priority flow channel signal, suppressing background noise interference, inputting the salt-wet coupling strength index into the trained random forest classifier, automatically segmenting the priority flow path region, and outputting the spatial vector boundary of the salt barrier failure region to achieve automatic segmentation of the failure region; It should be noted that the improved Normalized Differential Water Index (MNDWI) and Salinity Index (NDSI) raster are pixel-level normalized. A minimum-maximum normalization method is used to map the dual index values ​​to the [0, 1] interval, eliminating dimensional differences. After normalization, a dual-channel fused feature map is constructed, with the first channel being the normalized MNDWI and the second channel being the normalized NDSI. Each pixel corresponds to a two-dimensional feature vector. The spatial resolution of the fused feature map remains consistent with the original remote sensing image, with a pixel size of 10m × 10m, covering all candidate areas of salt barrier failure. This ensures that the two indices have equal weight contributions in subsequent principal component analysis, avoiding feature extraction bias due to dimensional differences. Principal component analysis is used to reduce the dimensionality of the dual-channel fused feature map, calculating the 2×2 covariance matrix and its eigenvalues ​​and eigenvectors. The first principal component with the larger contribution rate is extracted as the salt-wet coupling strength index. The loading coefficient of the first principal component reflects the contribution weight of MNDWI and NDSI to the coupling strength. The higher the value, the more significant the salt-wet coupling. The first principal component raster image is output as a single-channel grayscale image. In this image, the priority flow channel region appears as continuous bright stripes or a mesh structure, while the background region shows a low-value uniform distribution, effectively highlighting the spatial signal characteristics of the priority flow channel. The salt-wet coupling strength index raster image is input into a trained random forest classifier for classification and segmentation. The classifier contains 100 decision trees, with a maximum depth limit of 10 layers and a minimum number of leaf node samples of 5. The input features are the salt-wet coupling strength value of each pixel and its 8-neighborhood local variance, forming a 9-dimensional feature vector. The classifier outputs the posterior probability of each pixel belonging to the priority flow path. A probability threshold of 0.6 is set to extract the priority flow path pixel set. After smoothing the edges through morphological closing operations (structural element is 3×3 pixels), it is converted into a spatial vector boundary and outputs a Shapefile file of the salt barrier failure area, containing attribute fields: failure type (damaged / unevenly laid), area, and average salt-wet coupling strength. Step 3: Deploy stratified temperature, humidity, and salt sensors in the identified failure areas, simultaneously access real-time tidal water level data and groundwater depth monitoring information, establish a dynamic monitoring network for the salt return process, construct a multi-source dynamic sensing system, and support refined analysis of the salt return process. Deploy stratified temperature, humidity, and salt sensors in the identified salt barrier failure areas, with monitoring depths of at least 5cm, 15cm, 25cm and 40cm below the salt barrier layer, to acquire vertical profile data of soil temperature, humidity, and electrical conductivity, achieving accurate tracking and stratified quantification of the vertical migration path of salt. Simultaneously access real-time tidal water level data of the study area and the depth and mineralization information of adjacent groundwater monitoring wells, forming a multi-source data synchronous acquisition link of surface-groundwater-tidal, constructing a multi-factor collaborative observation system for the salt return process, improving mechanism analysis capabilities, setting the sensor data acquisition frequency to no less than once per hour, constructing a dynamic monitoring network for the salt return process covering the complete tidal cycle, to support the time-series analysis of the salt return mode, ensure continuous capture of the salt return dynamic throughout the entire cycle, and eliminate monitoring blind spots. It should be noted that within the identified areas of salt barrier failure, stratified temperature, humidity, and salt sensors were installed using a priority flow channel cross-sectional layout. A sensor assembly was vertically buried at each monitoring point, with monitoring depths set at 5cm, 15cm, 25cm, and 40cm below the salt barrier, corresponding to the surface salt accumulation zone, salt return peak zone, root activity zone, and control zone below the salt barrier. The sensor assembly consisted of an EC-5 conductivity probe, a 5TM temperature and humidity probe, and a data acquisition module. Each sensor was calibrated in the laboratory before on-site installation, followed by backfilling and compaction of the original soil to ensure close contact between the probe and the soil. Real-time tidal water level data for the study area was simultaneously accessed from a nearby tidal station, with a sampling interval of 6 minutes, transmitted via an IoT gateway using MQTT protocol. The data is transmitted to the local server, and simultaneously accesses groundwater monitoring well data within 200m of each monitoring point, including groundwater level depth and salinity. The sampling frequency is once per hour. Surface-ground-tidal multi-source data are synchronized using a unified timestamp to form a time-aligned multi-source dataset for subsequent salinization mechanism analysis. All sensor data are set to be collected once per hour, with continuous collection time of no less than 30 days, covering at least two complete tidal cycles (each cycle is approximately 14.8 days). After preliminary filtering and noise reduction by edge computing nodes, the collected data is uploaded to the central database to construct a dynamic monitoring network for the salinization process. This network outputs time-series curves of conductivity, temperature, and humidity at various depths, as well as synchronized tidal level and groundwater level fluctuation curves. Step 4: Use time-series clustering algorithm to separate capillary action and tidal backflow driven salt return modes, analyze the dominant salt return mechanism in different regions, clarify the failure cause type, quantitatively distinguish the salt return driving mechanism, and avoid blindly applying treatment measures. Step 5: Based on the classification results of the salt return dominant mechanism, construct a soil salinity change prediction model, integrate soil salinity dynamics, meteorological and tidal cycle data, output the remediation window period, realize quantitative prediction of remediation timing, and avoid unfavorable windows of tidal evaporation. Step 6: Combining the output repair window period, use reinforcement learning algorithm to optimize and output the optimal salt barrier layer replacement scheme, realize the accurate repair and functional restoration of the salt barrier layer failure area, form a closed-loop detection-decision-control system, and continuously improve the accuracy of repair decision-making.

[0022] Example 2, as Figure 1 , Figure 2 As shown, based on Example 1, the present invention provides a technical solution: Step 4 specifically includes: extracting soil conductivity time-series data at various depths for at least 15 consecutive days from the dynamic monitoring network of the salt return process, constructing a multi-dimensional time series sample set, each sample containing depth and time dimension information, realizing the complete preservation of the vertical dynamic characteristics of the salt return process, using the dynamic time warping algorithm to calculate the similarity distance between time series samples at different locations, eliminating the influence of phase shift caused by tides and evaporation on clustering, eliminating phase interference caused by tides and evaporation, improving clustering accuracy, and combining the K-means clustering algorithm to divide the time series samples into two salt return dominant mechanisms, including capillary action dominant type and tidal support dominant type, outputting the salt return type label of each failure area, realizing the automatic identification and classification of the salt return dominant mechanism; It should be noted that after constructing the dynamic monitoring network for the salt return process, soil electrical conductivity time-series data for more than 15 consecutive days at various depths were extracted, with a sampling interval of 1 time / hour. For each monitoring point, four depths (5cm, 15cm, 25cm, 40cm) of electrical conductivity time series were generated. These multi-depth series for each monitoring point were combined into a multi-dimensional sample vector with dimensions of 4×360 (15 days × 24 hours), constructing a time-series sample set covering all monitoring points in the failure area. Each sample in the set fully preserves the dynamic variation characteristics of soil salinity on the vertical profile, including the phase relationship and amplitude difference between the surface salt accumulation zone and the salt return peak zone. A dynamic time warping algorithm was used to calculate the similarity distance between the time-series samples of each monitoring point. This algorithm uses a non-linear bending time axis to elastically align similar feature points in two sequences, effectively eliminating the interference of tidal cycle phase difference (up to 6 hours) and diurnal evaporation rhythm shift on the similarity measurement. To avoid unreasonable alignment caused by overfitting, a curved window width of 72 hours was set. After calculating the regular distance between all monitoring points, a distance matrix was constructed as the clustering input to reflect the true similarity of each monitoring point in the dynamic waveform of salt return. The time series samples were divided into two salt return modes using the K-means clustering algorithm. The number of clusters was set to K=2. The initial cluster centers were the two sample points farthest apart in the distance matrix. During the iteration process, each sample was assigned a cluster center based on its dynamic time regular distance from the cluster center until the cluster center no longer changed. In the output clustering results, the first type was capillary-dominated, characterized by the salt peak appearing in the shallow layer (5cm or 15cm) and strongly correlated with evaporation and groundwater depth. The second type was tidal-dominated, characterized by the salt peak appearing in the 25cm or deeper layer and synchronizing with high tide events. Each monitoring point in the failure area was assigned a corresponding salt return type label. Furthermore, step 4 also includes: for areas determined to be dominated by capillary action, analyzing the temporal correlation between soil salinity changes and evaporation and groundwater depth, identifying the gradual salt return characteristics induced by uneven salt barrier layer laying, accurately distinguishing the causes of gradual salt return, and avoiding confusion with damage-induced salt return; for areas determined to be dominated by tidal backwater, calculating the phase difference and time lag coefficient between the peak salinity and high tide events, identifying the pulse-like salt return characteristics induced by local damage to the salt barrier layer, quickly locating the pulse-like salt return caused by damage, providing a basis for targeted remediation, and classifying the failure causes into uneven laying, damage, or tidal-induced types based on the comprehensive salt return pattern type and spatial distribution, serving as the basis for subsequent remediation strategy selection, realizing a closed loop of failure cause classification, and improving the matching degree between remediation strategies and on-site conditions; It should be noted that, for failure areas determined to be dominated by capillary action, time-series conductivity data at depths of 5cm and 15cm were extracted from each monitoring point. Simultaneously, daily evaporation from nearby meteorological stations and hourly water level depth from groundwater monitoring wells were acquired. Pearson correlation analysis was used to calculate the lag cross-correlation coefficient between conductivity changes and evaporation and water level depth. The time window was set to 72 hours with a step size of 6 hours. When the time lag between the peak conductivity and the peak evaporation was less than 12 hours, and the lag between the peak conductivity and the water level trough (the shallowest point of groundwater) was between 6 and 18 hours, the area was determined to be a salt barrier layer. The unevenness-induced gradual salt return characteristic is characterized by slow salt accumulation in shallow layers, fluctuation amplitude of less than 30%, and a salt return peak-to-valley ratio of less than 2.0, reflecting that uneven laying leads to enhanced capillary continuity rather than structural failure. For the failure area determined to be dominated by tidal backwater, conductivity time-series data at a depth of 25cm and 6-minute tidal level data from nearby tidal stations were extracted. The tidal level data were resampled to hourly averages, and cross-correlation analysis was used to calculate the phase difference between the conductivity peak and the high tide event (tidal level greater than the local average high tide level). The search window was set to the previous 6 hours to the following 12 hours, and the time lag coefficient was calculated. ,in The time lag between the peak conductivity and the high tide level. The tidal cycle is semi-diurnal (12.42 hours). When the conductivity value is between 0.15 and 0.35 (i.e., lag of 1.9 to 4.3 hours) and the increase in conductivity during a single salt return event is greater than 50%, it is identified as a pulsed salt return characteristic induced by local damage to the salt barrier. The preferential flow channel formed by the damage allows seawater to rapidly intrude under tidal backflow, and the salt content exhibits a peak pulse pattern. The duration of a single event is usually less than 6 hours. Based on the salt return pattern type and spatial distribution characteristics, the causes of each failure area are classified into three types: uneven laying type (capillary action dominates and gradual salt return), damage type (tidal backflow dominates and pulsed salt return), and tidal-induced type (tidal backflow dominates but there are no obvious signs of damage, and salt return is stably synchronized with high tide). The classification criteria include: cluster labels and time lag coefficient. Value range, increase and duration of salt return events, uneven paving type Undefined or greater than 0.5, broken type Between 0.15 and 0.35 with an increase of more than 50%, tidal-induced type. For values ​​between 0 and 0.15 and increases between 20% and 40%, output a Shapefile file. Each polygon attribute table contains a failure type field (1-uneven paving, 2-damage, 3-tidal induced) and a suggested repair measure code. Step 5 specifically includes: based on the classification results of the salt return dominant mechanism, constructing a multi-dimensional feature input set that integrates soil salinity dynamics, daily evaporation, precipitation and tidal cycle data to form a remediation decision sample library, establishing high-quality training samples to improve the model's prediction reliability and generalization ability, using a long short-term memory network to construct a soil salinity change prediction model, learning the nonlinear mapping relationship between salt return rate and meteorological-tidal factors, predicting the soil salinity change trend in the next 7 to 15 days, realizing early prediction of salt return trend, and gaining sufficient preparation time for remediation operations. Based on the condition that the predicted salt return is below the threshold and lasts for at least 72 hours, the optimal remediation window period is output. At the same time, the high tide and high evaporation periods that are not suitable for remediation are marked to accurately avoid unfavorable periods and significantly improve the success rate of salt barrier remediation operations. It should be noted that the input feature vector for each monitoring point includes the following dimensions: hourly conductivity values ​​at depths of 5cm, 15cm, and 25cm over the past 72 hours (216 in total), daily evaporation (from nearby meteorological stations), hourly precipitation, tidal cycle phase angle (normalized to 0~2π with a period of 12.42 hours), and the predicted tidal level for the next 24 hours (from tidal table forecasts). All features are mapped to the [0, 1] interval after minimum-maximum normalization and arranged chronologically with a 6-hour sliding step to form supervised learning sample pairs: the input is the feature sequence of the past 72 hours (73 time points, including the current time), and the output is the predicted conductivity values ​​at each depth for the next 72 hours. The repair decision sample library contains at least 2000 valid samples, covering three typical periods: spring salt return period, summer salt leaching period, and autumn stabilization period. The library is divided into training, validation, and test sets in a 7:1.5:1.5 ratio. The soil salinity change prediction model uses a three-layer stacked LSTM structure with 128, 64, and 32 neurons in the hidden layers, followed by a Dropout layer with a dropout rate of 0.3 to prevent overfitting. The input time step is set to 72 hours, and the feature dimension for each time step is 22 (3 depths of conductivity × 24 hours aggregated into 12 statistical features, including mean, maximum, minimum, and rate of change, combined with 10 meteorological tidal factors, and 22 effective features retained after feature selection). The output layer is a fully connected layer, and the output dimension corresponds to the hourly predicted values ​​of conductivity at each depth for the next 72 hours (a total of 3 depths × 72 hours = 216 output nodes). The model training uses mean squared error as the loss function, and the initial learning rate of the Adam optimizer is set to 0.001. The size is set to 32, the maximum training epochs are 200, and the early stopping mechanism is set to patience = 15 epochs. During training, a validation set is used for monitoring; training is terminated and the model rolls back to the optimal weights when the validation set loss does not decrease for 15 consecutive epochs. The model evaluation metrics are root mean square error (RMSE) and mean absolute percentage error (MAPE). For deployment in real-world prediction, RMSE ≤ 0.8 mS / cm and MAPE ≤ 12% are required on the test set. The salt return threshold is set based on the mean background conductivity, with a mild salt return threshold of [missing value]. Moderate Severe The repair window requires that the predicted conductivity at all depths be below the mild salinity threshold for 72 consecutive hours in the future, and there be no precipitation event greater than 10 mm within 48 hours from the start of the window (to avoid temporary salinity drop caused by scouring interfering with the judgment). The rule for determining the period unsuitable for repair is: the tidal level continuously exceeds the local average high tide level within the next 48 hours and the daily evaporation is greater than 5 mm / d during the same period. Under this condition, capillary salinity return and tidal backwater are superimposed, and the salinity reduction measures of the repair operation are difficult to maintain the effect. The model is run once a day and outputs the daily repair suitability score for the next 15 days. The score ≥80 points is marked as the recommended window period, and the score ≤30 points is marked as the period unsuitable for repair. The results are pushed to the construction management terminal in the form of a time axis color band chart to guide the on-site repair operation scheduling. The expression for the daily repair suitability score is as follows: ; In the formula: For the first The day is assessed for suitability for remediation, with higher scores indicating that the day is more suitable for salt barrier remediation work. For the first The maximum value of soil electrical conductivity at each depth (5cm, 15cm, 25cm) predicted by the model within the next 72 consecutive hours; The threshold for mild salt return is calculated using the following formula: ; The threshold for severe salt return is calculated using the following formula: Used to normalize the degree of conductivity exceeding the standard. The average conductivity of the background field. The standard deviation of the background field conductivity; For the first The maximum hourly rainfall in the next 48 hours starting today; The precipitation disturbance threshold is set at 10 mm. When the precipitation is greater than 10 mm, it is considered that precipitation erosion will cause a temporary decrease in soil salinity, which will interfere with the judgment of the true level of salinity return and thus reduce the suitability for remediation. For the first The cumulative number of days with a daily evaporation rate greater than 5 mm / d within the next 48 hours starting from today; The threshold number of days for the superposition of tide and evaporation is set to 1 day. When the evaporation is greater than 5 mm / d and the number of days that the tide level exceeds the local average high tide level exceeds this value, it is determined to be a condition that is not suitable for repair. The maximum number of unfavorable days is set to 2 days (i.e., the upper limit of the next 48 hours). Step 6 specifically includes: taking the spatial distribution of the salt barrier failure area, the failure cause type, and the repair window period as the state space input, defining the repair location, depth, thickness, and material type as the action space, realizing the quantitative representation and decision modeling of the multi-attribute state of the failure area, constructing a multi-objective reward function with the goal of minimizing the salt return rate after repair and maximizing the salt barrier function recovery period, using the Q-learning algorithm to iteratively train the repair strategy, balancing salt suppression and functional durability, improving the overall performance of the strategy, and outputting the optimal salt barrier repair scheme, including the repair area spatial vector, layered structure parameters, and drainage pipe adjustment suggestions, to guide precise on-site repair construction, generate a refined repair scheme that can be directly constructed, and improve on-site operation efficiency; It should be noted that after completing the identification of failure areas, classification of causes, and prediction of repair window periods, the spatial distribution vector of the salt barrier failure areas, the failure cause type codes (1-uneven laying type, 2-damage type, 3-tidal induced type), and the daily repair suitability score for the next 15 days are used as the state space input. The state vector dimension is the number of failure areas × (spatial coordinates, area, type code, 15-day score). The action space is defined as a discrete combination of replenishment parameters: the replenishment location is selected within a 2.5m buffer zone from the center point of the failure area; the replenishment depth is set to 5cm or 10cm below the original salt barrier; the replenishment thickness is set to 10cm, 15cm, and 20cm; the material... The types selected are graded crushed stone (particle size 5-25mm) or a composite layer of crushed stone and geotextile. The total number of action combinations is 2×3×2=12. An ε-greedy strategy (initial ε=0.3) is used for action exploration. Each iteration starts from the current state and selects a set of supplementary parameters to execute. The multi-objective reward function focuses on minimizing the salt return rate after repair and maximizing the salt barrier function recovery period. The salt return rate after repair is defined as the ratio of the maximum conductivity at a depth of 15cm in the failure area to the mild salt return threshold (μ+0.5σ) 30 days after the repair is completed. When it is less than 1, the return rate is 0. The salt barrier function recovery period is defined as the period from the repair to the first increase in conductivity at a depth of 15cm. The reward function is calculated as follows: R = 0.6 × (1 - return rate) + 0.4 × (recovery period / 90), where the upper limit of the recovery period is 90 days. The learning rate α of the Q-learning algorithm is set to 0.1, the discount factor γ is set to 0.9, and 500 rounds of iterative training are performed. In each round of training, the Q-value table is updated based on the current state-action pair. The training convergence condition is that the optimal action remains unchanged for 50 consecutive rounds. After training, the optimal action combination for each state is extracted to form a replenishment strategy decision table. Based on the Q-value table after training convergence, the optimal salt barrier replenishment scheme is output for the state input of each failure region. The scheme is expressed in Sh... Output the spatial vector boundary of the supplementary area in APEfile format. The boundary is obtained by expanding the polygon of the failed area by 1m to ensure coverage of all priority flow channels. The layered structural parameters include: the bottom salt barrier layer is made of graded crushed stone with a thickness of 10-20cm (selected according to the original failure type: 15cm for uneven laying, 20cm for damaged type, and 10cm for tidal induced type), the middle layer is laid with a layer of polypropylene geotextile (specification 200g / m²), and the top is backfilled with 30cm of original soil mixed with 3% straw fragments. The drainage pipe adjustment suggestion is: add DN75 perforated corrugated pipes along the long side of the supplementary area, buried 5cm above the salt barrier layer, with a spacing of no more than 6m and a slope of no less than 0.3%, the outlet is connected to the existing drainage well, and the plan document also includes a construction sequence description, including excavation, foundation cleaning, gravel laying and compaction, geotextile overlapping (overlap width not less than 20cm), backfilling and underground pipe installation requirements, which are directly pushed to the on-site construction terminal for precise repair work; Soil salinization was classified according to the "Soil Salinization Classification Standard" (SL 583-2012). Based on the actual conditions of the study area, soil total salt content was used as the core evaluation index, and the soil salinization was divided into 5 levels. The specific classification standards are shown in Table 1. Table 1. Classification Standards for Soil Salinization Furthermore, step 6 also includes: after the salt barrier layer is repaired, reactivating the distributed soil conductivity sensor array, collecting spatial distribution data of soil conductivity above the repaired salt barrier layer, realizing rapid and accurate verification of salt distribution after repair, comparing the difference between the repaired data and the initial background field, calculating the salt reduction rate and spatial uniformity improvement of each failed area, evaluating the repair effect, quantitatively determining the repair level and identifying residual failure risks, feeding back the repair effect evaluation results to the soil salinity change prediction model, updating the sample library and model parameters, realizing closed-loop iteration and adaptive optimization of the entire process of detection, decision-making and control, and continuously improving the model prediction accuracy and the adaptability of the repair strategy; It should be noted that after the salt barrier layer remediation is completed, the distributed soil conductivity sensor array, arranged in a 5m×5m grid, is reactivated. Each four-ring probe maintains a measurement depth of 0-30cm. Under conditions of no rainfall and no artificial irrigation for three consecutive days, each measuring point is precisely located using RTK-GPS, and three consecutive apparent conductivity values ​​are collected and their arithmetic mean is taken. The collected remediation dataset is then compared with the initial background field established before construction, performing pixel-by-pixel difference calculations to generate a conductivity change raster map. The raster pixel size remains at 0.5m×0.5m. Based on the raster, the salinity reduction rate is calculated for each identified failure area. The formula for calculating the reduction rate is... The formula is (mean conductivity before remediation minus mean conductivity after remediation) divided by the mean conductivity before remediation. Simultaneously, the change in the coefficient of variation of conductivity within the region is calculated to characterize the degree of improvement in spatial uniformity. A salinity reduction rate greater than 70% and a coefficient of variation reduction of more than 30% are considered excellent remediation; a reduction rate between 40% and 70% is considered acceptable; and a rate below 40% requires a new round of testing. The remediation effect evaluation results are fed back to the soil salinity change prediction model in structured data form. The evaluation results include the salinity reduction rate of each affected area, the degree of improvement in spatial uniformity, and salinity return tracking data within 30 days after remediation. After receiving the feedback, the model will add the new data... Comparative samples before and after restoration, along with time-series data on salt return tracking, were included in the training sample library. The library covers three typical periods: spring salt return, summer salt leaching, and autumn stabilization. Model parameters were updated using incremental learning, maintaining a three-layer stacked LSTM architecture (128, 64, and 32 hidden layer neurons). The actual observed salt return process after restoration was used as the validation benchmark, and model weights were fine-tuned, focusing on correcting the nonlinear mapping between salt return rate and meteorological-tidal factors. The updated model significantly improved the accuracy of salt change predictions for similar failure areas in subsequent rolling forecasts. This mechanism constructs a closed-loop adaptive system from detection to decision-making and then to regulation. The optimized system uses the evaluation results of each repair operation as prior knowledge for the next decision, driving continuous optimization of the state space and action space. The dynamic parameters in the state space are updated with the feedback of the repair effect, and the parameters in the action space, such as the repair thickness and material selection, are adjusted according to the actual repair effect. The Q-value table is updated after each closed-loop iteration. The learning rate α is kept at 0.1, the discount factor γ is kept at 0.9, and the convergence condition is maintained at the standard that the optimal action has not changed for 50 consecutive rounds. After multiple closed-loop iterations, it can automatically identify the optimal repair strategy combination for different failure types, and achieve precise control and long-term stable maintenance of the salt barrier function restoration.

[0023] The above are merely specific embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. The scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for detecting and analyzing the salt leaching effect of a salt-barrier layer in severely saline coastal soil, characterized in that, Includes the following steps: Step 1: Based on the intelligent sensing system deployed in the study area, the distributed soil conductivity sensing array is used to obtain the spatial distribution of soil conductivity above the salt barrier layer, construct the initial background field of salt leaching effect, and locate potential failure areas. Step 2: Combining the improved spatial fusion algorithm of normalized differential water index and salinity index, the image features of the priority flow path are located by image recognition to determine the area of ​​salt barrier failure, that is, the area of ​​local damage or uneven laying. Step 3: Deploy layered temperature, humidity and salinity sensors in the identified failure areas, and simultaneously access real-time tidal water level data and groundwater depth monitoring information to establish a dynamic monitoring network for the salinization process. Step 4: Use temporal clustering algorithm to separate the salt return modes driven by capillary action and tidal back pressure, analyze the dominant salt return mechanism in different regions, and clarify the failure cause type; Step 5: Based on the classification results of the salt return dominant mechanism, construct a soil salinity change prediction model, integrate soil salinity dynamics, meteorological and tidal cycle data, and output the remediation window period; Step 6: Based on the output repair window period, use reinforcement learning algorithm to optimize and output the optimal salt barrier layer replacement scheme to achieve accurate repair and functional restoration of the salt barrier layer failure area.

2. The method for detecting and analyzing the salt leaching effect of salt-isolating layer deployment in severely saline coastal soil according to claim 1, characterized in that: Step 1 specifically includes: A smart sensing system was deployed in a 5m×5m grid within the study area. The distributed soil conductivity sensor array included in the smart sensing system was used to collect apparent conductivity data of the soil layer 0-30cm above the salt barrier layer, thereby obtaining spatially continuous conductivity distribution information. The ordinary Kriging interpolation method was used to spatially reconstruct the collected soil apparent electrical conductivity data to generate an initial background field for salt leaching effect. The mean and standard deviation of the background field were calculated to identify areas with abnormally high electrical conductivity values. Using the background mean plus 1.5 times the standard deviation as a threshold, continuous spatial units above the threshold are extracted and preliminarily identified as potential failure zones of the salt barrier layer.

3. The method for detecting and analyzing the salt leaching effect of salt-isolating layer deployment in severely saline coastal soil according to claim 1, characterized in that: Step 2 specifically includes: High-resolution multispectral remote sensing images of the study area were acquired, the improved normalized differential water index was calculated, and the spatial distribution of the surface wet zone was extracted as a water indicator feature of the priority flow path. Simultaneously calculate the salinity index, extract the spatial distribution of soil salinity enrichment areas, and spatially overlay it with the distribution of the humid zone to identify anomalous salt-humidity coupling patches. A semantic segmentation network was used to perform image recognition on superimposed salt-wet coupling abnormal patches, extract the preferential flow path image features of strip or network distribution, and determine the local damage or uneven laying area of ​​the salt barrier layer.

4. The method for detecting and analyzing the salt leaching effect of salt-isolating layer deployment in severely saline coastal soil according to claim 3, characterized in that: Step 2 also includes: The improved normalized differential water index and salinity index are normalized at the pixel level to eliminate dimensional differences and generate a dual-channel fused feature map. Principal component analysis was used to reduce the dimensionality and enhance the features of the dual-channel fused feature map. The first principal component was extracted as the salt-wet coupling strength index to highlight the spatial signal of the preferred flow channel. The salt-wet coupling strength index is input into the trained random forest classifier to automatically segment the priority flow path region and output the spatial vector boundary of the salt barrier failure region.

5. The method for detecting and analyzing the salt leaching effect of salt-isolating layer deployment in severely saline coastal soil according to claim 1, characterized in that: Step 3 specifically includes: Layered temperature, humidity and salt sensors were deployed in the identified salt barrier failure area, with monitoring depths of at least 5cm, 15cm, 25cm and 40cm below the salt barrier layer, to obtain vertical profile data of soil temperature, humidity and electrical conductivity. Simultaneously access real-time tidal water level data of the study area and burial depth and mineralization information of nearby groundwater monitoring wells to form a multi-source synchronous acquisition link of surface-groundwater-tidal data; Set the sensor data acquisition frequency to no less than once per hour, and construct a dynamic monitoring network covering the salt return process throughout the entire tidal cycle.

6. The method for detecting and analyzing the salt leaching effect of salt-isolating layer deployment in severely saline coastal soil according to claim 1, characterized in that: Step 4 specifically includes: Extract time-series data of soil electrical conductivity at various depths for at least 15 consecutive days from the dynamic monitoring network of the salt return process, and construct a multidimensional time series sample set, with each sample containing depth and time dimension information; A dynamic time warping algorithm is used to calculate the similarity distance between time series samples at different locations, eliminating the impact of phase shift caused by tides and evaporation on clustering; By combining the K-means clustering algorithm, the time series samples are divided into two types of salt return patterns: capillary action-dominated and tidal support-dominated. The salt return type label of each failure region is output.

7. The method for detecting and analyzing the salt leaching effect of salt-isolating layer deployment in severely saline coastal soil according to claim 6, characterized in that: Step 4 also includes: For areas identified as being dominated by capillary action, the temporal correlation between soil salinity changes and evaporation and groundwater depth was analyzed to identify the gradual salt return characteristics induced by uneven salt barrier layer laying. For areas identified as being dominated by tidal backwater, the phase difference and time delay coefficient between the peak salinity and the high tide event are calculated to identify the pulse-like salt return characteristics induced by local damage to the salt barrier layer. Based on the combined salt return pattern type and spatial distribution, the causes of failure are classified into uneven laying type, damage type, or tidal induced type.

8. The method for detecting and analyzing the salt leaching effect of salt-isolating layer deployment in severely saline coastal soil according to claim 1, characterized in that: Step 5 specifically includes: Based on the classification results of the dominant mechanism of salt return, a multi-dimensional feature input set integrating soil salinity dynamics, daily evaporation, precipitation and tidal cycle data is constructed to form a sample library for remediation decisions; A soil salinity change prediction model was constructed using a long short-term memory network. The nonlinear mapping relationship between salt return rate and meteorological-tidal factors was learned to predict the soil salinity change trend in the next 7 to 15 days. Given the condition that the predicted salinity return is below the threshold and lasts for at least 72 hours, the optimal repair window is output, while the high tide and high evaporation periods that are not suitable for repair are marked.

9. The method for detecting and analyzing the salt leaching effect of salt-isolating layer deployment in severely saline coastal soil according to claim 1, characterized in that: Step 6 specifically includes: The spatial distribution of the salt barrier failure area, the failure cause type, and the repair window period are used as the state space input, and the replacement location, depth, thickness, and material type are defined as the action space. A multi-objective reward function is constructed with the goal of minimizing the salt return rate after repair and maximizing the salt isolation function recovery period. The Q-learning algorithm is used to iteratively train the repair strategy. Output the optimal salt barrier layer replacement scheme, including the spatial vector of the replacement area, layer structure parameters, and drainage pipe adjustment suggestions, to guide precise on-site repair construction.

10. The method for detecting and analyzing the salt leaching effect of salt-isolating layer deployment in severely saline coastal soil according to claim 9, characterized in that: Step 6 also includes: After the salt barrier layer repair is completed, the distributed soil conductivity sensor array is reactivated to collect spatial distribution data of soil conductivity above the repaired salt barrier layer. The difference analysis was performed between the repaired data and the initial background field to calculate the salt reduction rate and the degree of improvement in spatial uniformity of each failed area, and to evaluate the repair effect. The results of the remediation effect assessment are fed back to the soil salinity change prediction model to update the sample library and model parameters, realizing closed-loop iteration and adaptive optimization of the entire process of detection, decision-making and regulation.

Images