Search

CN-121982381-A - Pixel-level identification and objectification semiautomatic classification integration method for frozen and thawed slope products in alpine mountain area

CN121982381ACN 121982381 ACN121982381 ACN 121982381ACN-121982381-A

Abstract

The invention discloses a pixel-level identification and objectification semiautomatic classification integration method for freezing and thawing slope products in alpine mountainous areas, and relates to the technical field of remote sensing and geographic information systems. Under the complex topography of alpine mountain areas and strong seasonal background, freeze-thawing slope products are stably and accurately extracted on a long time sequence scale by utilizing multi-source remote sensing data, pixel-level results are converted into plaque objects available for engineering, and under the condition of lacking year-by-year complete environment covariates, the three-dimensional interpretable semi-automatic classification of mechanism-space distribution-substance composition is realized, and reproducible and renewable space-time evolution databases and map products are finally output.

Inventors

  • YE HUI
  • LIU SHIYIN
  • WANG JINLIANG
  • BAI DIE

Assignees

  • 云南师范大学

Dates

Publication Date
20260505
Application Date
20260114

Claims (10)

  1. 1. A pixel-level identification and objectification semiautomatic classification integration method for frozen and thawed slope products in alpine mountain areas is characterized by comprising the following steps: The data acquisition and multisource feature fusion comprises the steps of acquiring long-time-sequence multisource remote sensing data and topographic data of a target area, performing radiation correction, topographic correction, cloud and shadow masking and annual synthesis pretreatment, and fusing optical spectrum, texture, surface heat environment, freeze thawing process indication and topographic factors to construct a multisource feature stack for freeze thawing slope product identification; processing the multi-source feature stack by using a supervision classification model to generate a pixel-level probability distribution map of the freeze-thawing slope product, and binarizing the probability distribution map into an initial pixel-level distribution mask by a preset probability threshold; Carrying out morphological filtering on the pixel component dividing mask to remove noise and tiny broken plaques, and dividing the pixel component dividing mask into independent and continuous planar plaque objects through connected domain analysis; The method comprises the steps of carrying out primary judgment on three mutually orthogonal dimensions of a forming mechanism, spatial distribution and substance composition based on a preset knowledge rule base by utilizing engineering attributes of plaque objects, outputting a primary judgment result and confidence level thereof, calling an object-level machine learning model to carry out secondary judgment on plaque objects with conflict or confidence level lower than a preset threshold value on the primary judgment result, merging the primary judgment result of the knowledge rule and the judgment result of the machine learning model to generate a final three-dimensional classification label of each plaque object, marking few plaques with merged classification certainty still lower than the requirement and sending the few plaques into a manual review queue, and feeding back the review result to optimize the threshold value and model parameters; And (3) detecting and outputting the space-time evolution of the multi-period three-dimensional classification result, namely carrying out space-time alignment and consistency processing on plaque object data sets which are generated in different periods and contain three-dimensional classification labels, identifying and counting the dynamic change of the freezing and thawing slope products in area, position and type through a post-classification comparison method, and outputting a space-time evolution drawing and a database of the space-time evolution drawing.
  2. 2. The method for pixel-level identification and objectification semiautomatic classification integration of frozen and thawed slope products in alpine mountainous areas according to claim 1, wherein the construction of the multi-source feature stack specifically comprises the following steps: the freeze thawing state or soil moisture product obtained by combining passive microwave data inversion is used as a freeze thawing process indication characteristic to enhance the identification capability of the freeze thawing process sensitive period, and the spectral indexes at least comprise normalized vegetation indexes NDVI and normalized snow index NDSI, and the calculation formulas are as follows: , , Wherein, the Is the surface reflectivity of the corresponding wave band.
  3. 3. The method for semi-automatic classification integration of pixel level identification and objectification of freeze-thawing alluvial substances in alpine mountain areas according to claim 1, wherein the supervised classification model is a random forest model, and the probability that pixels in the freeze-thawing alluvial substance probability map belong to the freeze-thawing alluvial substance category is characterized in that The calculation formula is as follows: , Wherein, the For the number of decision trees in a random forest, Is the first The discriminant function of the decision tree, Is an indication function; Presetting a probability threshold Optimal selection based on F1 score or Youden index of verification set, binary mask The generation method of (1) comprises the following steps: , Wherein, the To obtain a freeze-thaw-slope-product probability map.
  4. 4. The method for classifying and integrating frozen and thawed slope products in alpine mountain area at pixel level according to claim 1, wherein the morphological filtering comprises open operation and closed operation, specifically comprising the steps of firstly adopting structural elements For binary mask Performing an open operation Then execute the closing operation To eliminate salt and pepper noise and smooth the boundary; The generation of the target plaque specifically comprises the steps of carrying out 8-neighborhood connected domain analysis on a filtered binary mask, marking each connected domain to form an initial plaque object set, setting a minimum drawing unit area threshold value The proposed area is smaller than Or merge it with an adjacent plaque.
  5. 5. The method for semi-automatic classification and integration of pixel level identification and objectification of frozen and thawed slope products in alpine mountain areas according to claim 1, wherein the classification rules of the mechanism dimension in initial judgment in three mutually orthogonal dimensions of formation mechanism, spatial distribution and material composition comprise: if the average gradient of plaque object And average vegetation coverage Judging that the materials are piled up in situ; If the average gradient of the plaque object is satisfied And Euclidean distance from nearest water system Judging that the slope toe is piled up; euclidean distance from nearest water system of plaque object And average gradient And judging that the valleys are accumulated.
  6. 6. The method for semi-automatic classification integration of pixel level identification and objectification of frozen and thawed slope products in alpine mountain areas according to claim 1, wherein the method for generating final three-dimensional classification labels of each plaque object by fusing initial judgment results of knowledge rules and judgment results of machine learning models is specifically as follows: preliminary classification is carried out on the plaque objects based on preset knowledge rules to obtain rule classification results A corresponding rule confidence; for rule classification results, conflicts exist or the confidence of the rules is lower than a preset threshold The plaque object of (2) is called an object-level gradient lifting model trained based on object attributes to carry out secondary classification, and model classification probability is obtained ; Determining final classification labels by means of weighted fusion The calculation formula is as follows: , , Wherein, the And Is a weight coefficient, and , Is an indication function; For the highest score after fusion The difference from the next highest score is less than a preset difference threshold The plaque object marked as the object to be checked is submitted to the manual checking queue.
  7. 7. The method for semi-automatic classification and integration of pixel level identification and objectification of freeze-thawing slope products in alpine mountain areas according to claim 5 or 6, wherein when classification rules forming mechanism dimensions are applied, a Sigmoid soft membership function is introduced to calculate membership degrees of three categories of plaque object belonging to in-situ accumulation, toe accumulation and valley accumulation for continuous discrimination thresholds of gradient, vegetation coverage and distance from water system The category corresponding to the maximum membership is used as a preliminary result of rule classification, and the maximum membership value is used as an initial value of rule confidence.
  8. 8. The method for pixel-level recognition and objectification semiautomatic classification integration of cold mountain area-oriented freeze-thaw slope products according to claim 1, wherein the calculated plaque object attributes comprise: Morphological properties-area Circumference of a circle Degree of tightness ; Terrain attribute average slope Average elevation Topography relief; hydrologic property-distance from water system ; Neighborhood attribute, distance from road 。
  9. 9. A computer readable storage medium having stored thereon a computer program, which when executed by a processor implements a method for pel level identification and targeting of freeze-thaw-slope products in alpine mountains as claimed in any one of claims 1-8.
  10. 10. An electronic device comprising a memory and a processor, the memory storing a computer program, wherein the processor, when executing the computer program, implements a pixel-level identification and targeting semi-automatic classification integration method for freeze-thaw-slope products in alpine mountains as claimed in any one of claims 1-8.

Description

Pixel-level identification and objectification semiautomatic classification integration method for frozen and thawed slope products in alpine mountain area Technical Field The invention relates to the technical field of remote sensing and geographic information systems, in particular to an integration method based on multi-source remote sensing data fusion and oriented to recognition, classification and dynamic monitoring of freezing and thawing slope products in alpine mountainous areas. The method is suitable for the fields of geological disaster risk assessment, engineering site selection, safety protection and the like. Background The alpine mountain area is subjected to the combined action of natural conditions such as high altitude, strong radiation, large day-night temperature difference, seasonality/multi-year frozen soil distribution and the like, the freezing and thawing cycle is frequent and high in strength, and a loose accumulation body, namely a freezing and thawing slope accumulation body, which is easy to form and continuously evolve under the long-term coupling action of frost heaving-thawing-gravity conveying-thawing/rainfall runoff redistribution of a slope rock-soil body. The pile is often concentrated in material source sensitive areas such as glacier retraction belt front edge, high altitude abrupt slope, canyon side slope, traffic engineering disturbance belt and the like, is not only an important material source and an amplifier of disaster chains such as landslide, debris flow, collapse-damming and the like, but also a key constraint factor of traffic infrastructure line selection, slope protection, engineering operation and maintenance and safety management of residential areas along the line. With the ice and snow melting, the space-time variability of precipitation and the frequency change of extreme events under the climate warming and humidification background in recent years, the slope material supply and aggregation process in the alpine region presents more remarkable stepwise fluctuation and spatial heterogeneity, and the freeze thawing slope product is subjected to large-scale, long-time sequence and reproducible fine drawing and dynamic monitoring, so that the method has become urgent technical requirements in geological disaster risk assessment, regional management and major engineering safety guarantee. The existing freeze thawing slope product investigation and drawing method still takes manual interpretation, field investigation and experience interpretation of a small amount of time phase images as main materials, and has the problems of low efficiency, high cost, large field risk, obvious limitation by weather and terrain shielding, difficulty in covering continuous corridor areas of mountain canyons and the like, and difficulty in supporting the systematic evaluation and annual update of transcounty domains/cross-domains. Even if the remote sensing automatic/semiautomatic method is adopted, the technical bottlenecks are generally faced with, firstly, that freeze thawing slope products are highly similar to bare rock, flood accumulating fans, sloughing accumulating, sparse vegetation and other ground products in spectrum-texture characteristics, single optical data or simple index threshold values are easy to generate common-spectrum foreign matters/common-matter different-spectrum confusion, secondly, shadow effect, slope radiation difference, seasonal performance change caused by atmospheric scattering and snow-snow melting alternation of complex topography are caused, so that stability of classification models is insufficient across areas, seasons and sensors, cross-year consistency and boundary comparability are difficult to ensure, thirdly, the existing flow always carries out split treatment on 'pixel-level identification' and 'planar plaque/object-level classification', the former is easy to generate salt-pepper noise and boundary fragmentation, the latter is lack of standardized classification rules for forming mechanism, space zoning, substance composition and other engineering interpretable properties, usability of achievements in disaster expression and engineering application is limited, and fourthly, multi-source data fusion, parameter management, quality recording and quality control and different regional risk of repeated achievement models are difficult to popularize, and different standards are difficult to cause. Along with the rapid development of multi-source remote sensing (optics, thermal infrared, microwaves, topography and the like), cloud computing platforms and geospatial big data technologies, long-time sequence pixel level identification is realized by means of mass influence data and parallel computing capability of Google EARTH ENGINE, and a ArcPy/ArcGIS targeted space analysis and rule-model collaborative discrimination mechanism is combined, so that a new feasible path is provided for constructing an integrated technology system of fr