CN-122024070-A - Mangrove growth condition monitoring method and system based on deep learning

CN122024070ACN 122024070 ACN122024070 ACN 122024070ACN-122024070-A

Abstract

The invention discloses a mangrove growth condition monitoring method and system based on deep learning, wherein the method comprises the steps of obtaining a double-time-phase remote sensing image of a target mangrove growth area, extracting a mangrove candidate growth area, and extracting deep semantic features to obtain mangrove structured growth change features; based on a preset evolution mode feature set, feature matching is carried out to obtain a mangrove evolution type probability distribution vector, a ternary feature sequence is constructed and input into a growth change rationality evaluation unit to carry out rationality evaluation to obtain high-confidence-degree features, and mangrove growth condition evaluation and regional mangrove growth condition index calculation are carried out to realize the monitoring of the mangrove growth condition. The invention can identify the mangrove range dynamic, evaluate the internal growth condition change and automatically generate the mangrove hierarchical management proposal. The mangrove growth condition monitoring method and system based on deep learning can be widely applied to the technical field of remote sensing image processing.

Inventors

DONG DI
TIAN SONG
GAO QING
LI XUERUI
GUO BINGXIN
HUANG HUAMEI
WEI ZHENG
SUN YUCHAO
JIANG LIN
LI KANG
YANG LEI
ZHANG XIAOHAO

Assignees

自然资源部南海发展研究院(自然资源部南海遥感技术应用中心)
自然资源部南海调查中心

Dates

Publication Date: 20260512
Application Date: 20260320

Claims (10)

1. The mangrove growth condition monitoring method based on deep learning is characterized by comprising the following steps of: Acquiring a first time phase remote sensing image and a second time phase remote sensing image of a target mangrove growth area, and performing mangrove candidate growth area extraction processing and double time phase depth semantic feature extraction processing to obtain mangrove structural growth change features; Based on a preset evolution mode feature set, feature matching is carried out on the mangrove structured growth change features to obtain a mangrove evolution type probability distribution vector; constructing a ternary characteristic sequence and inputting the ternary characteristic sequence into a growth change rationality evaluation unit for rationality evaluation to obtain high-confidence characteristics; Based on the high confidence characteristic and the mangrove evolution type probability distribution vector, mangrove growth condition assessment and regional mangrove growth condition index calculation are carried out, and the mangrove growth condition is monitored.
2. The method for monitoring the growth condition of mangrove forest based on deep learning according to claim 1, wherein the steps of obtaining a first time-phase remote sensing image and a second time-phase remote sensing image of a target mangrove forest growth area, and performing mangrove forest candidate growth area extraction processing and double time-phase deep semantic feature extraction processing to obtain mangrove forest structured growth variation features comprise the following steps: Acquiring a first time-phase remote sensing image and a second time-phase remote sensing image of a target mangrove growth area, and performing image preprocessing to obtain a preprocessed first time-phase remote sensing image and a preprocessed second time-phase remote sensing image; Respectively calculating vegetation characteristic indexes of the preprocessed first time-phase remote sensing image and the preprocessed second time-phase remote sensing image, setting a vegetation index threshold value, and performing non-vegetation region rejection processing to obtain a rejected first time-phase remote sensing image and a rejected second time-phase remote sensing image; Carrying out space constraint on the removed first time-phase remote sensing image and the removed second time-phase remote sensing image by combining with coastal zone geographic information to obtain a constrained first time-phase remote sensing image and a constrained second time-phase remote sensing image; combining the constrained first time-phase remote sensing image and the constrained second time-phase remote sensing image, manufacturing first time-phase mangrove forest distribution data and second time-phase mangrove forest distribution data through a random forest classification method, combining the first time-phase mangrove forest distribution data and the second time-phase mangrove forest distribution data, and generating a mangrove forest candidate growth area mask; Cutting the preprocessed first time-phase remote sensing image and the preprocessed second time-phase remote sensing image through a mangrove candidate growing area mask to obtain a cut first time-phase remote sensing image and a cut second time-phase remote sensing image; and based on the deep neural network, performing double-time-phase deep semantic feature extraction processing on the cut first-time-phase remote sensing image and the cut second-time-phase remote sensing image to obtain mangrove structured growth change features.
3. The method for monitoring the growth condition of the mangrove forest based on deep learning according to claim 2, wherein the step of obtaining the structural growth variation characteristic of the mangrove forest by performing double-time-phase deep semantic feature extraction processing on the cut first time-phase remote sensing image and the cut second time-phase remote sensing image based on the deep neural network specifically comprises the following steps: based on the visual feature extraction network, performing depth semantic feature extraction processing on the cut first time-phase remote sensing image and the cut second time-phase remote sensing image to obtain depth semantic features of the first time-phase remote sensing image and depth semantic features of the second time-phase remote sensing image; constructing a growth change feature based on the depth semantic feature of the first time-phase remote sensing image and the depth semantic feature of the second time-phase remote sensing image, wherein the growth change feature comprises a growth direction component, a growth intensity component and a structural consistency component; and performing channel splicing and feature fusion treatment on the three components of the growth change feature to obtain the mangrove structured growth change feature.
4. The method for monitoring the growth condition of the mangrove forest based on deep learning according to claim 3, wherein the step of performing feature matching on the structural growth change features of the mangrove forest based on the feature set of the preset evolution mode to obtain the probability distribution vector of the evolution type of the mangrove forest specifically comprises the following steps: constructing a preset evolution mode feature set, wherein the preset evolution mode feature set comprises a flourishing health type, an expanding new-born type, a damaged degradation type and an atrophy crushing type; Respectively carrying out L2 normalization calculation on a preset evolution mode feature set and mangrove structured growth change features to obtain normalized evolution modes and normalized feature vectors; Performing dot product calculation on the normalized evolution mode and the normalized feature vector to obtain cosine similarity of the evolution mode and the feature vector; By passing through And carrying out probability distribution conversion processing on the cosine similarity of the evolution mode and the feature vector by the function to obtain the mangrove evolution type probability distribution vector.
5. The method for monitoring the growth condition of mangrove forest based on deep learning according to claim 4, wherein the step of constructing a ternary feature sequence and inputting the ternary feature sequence to a growth change rationality evaluation unit for rationality evaluation to obtain a high confidence feature specifically comprises the following steps: sequentially carrying out dimension unification and splicing treatment on the depth semantic features of the first time-phase remote sensing image, the depth semantic features of the second time-phase remote sensing image and the mangrove structured growth change features to construct a ternary feature sequence; Building a growth change rationality assessment unit, wherein the growth change rationality assessment unit comprises an input mapping module, a local mode extraction module, a time sequence associated modeling module and an output mapping module; mapping the ternary feature sequence to a high-dimensional space based on an input mapping module to obtain a mapped ternary feature sequence; based on a local mode extraction module, carrying out local feature capturing processing on the mapped ternary feature sequence to obtain a ternary local feature sequence; Based on a time sequence associated modeling module, generating dynamic modulation weights, and carrying out characteristic state self-adaptive adjustment processing on the ternary local characteristic sequences to obtain adjusted ternary characteristic sequences; And based on the output mapping module, mapping and outputting the adjusted ternary feature sequence to obtain the high-confidence feature.
6. The method for monitoring the growth condition of the mangrove forest based on deep learning according to claim 5, wherein the steps of estimating the growth condition of the mangrove forest and calculating the regional mangrove forest growth condition index based on the high confidence characteristic and the probability distribution vector of the evolution type of the mangrove forest are realized, and the method specifically comprises the following steps: Fusing the high confidence coefficient characteristics with the mangrove evolution type probability distribution vector, evaluating the mangrove growth condition, outputting the mangrove growth condition score, and generating a mangrove quality degradation area mask, a mangrove area reduction area mask and a change type map; Based on the mangrove growth condition score and the mangrove evolution type probability distribution vector, calculating the regional mangrove growth condition index, and generating a mangrove hierarchical management suggestion to realize the monitoring of the mangrove growth condition.
7. The method for monitoring the growth condition of a mangrove forest based on deep learning as set forth in claim 6, wherein the step of merging the high confidence feature with the probability distribution vector of the evolution type of the mangrove forest, evaluating the growth condition of the mangrove forest, outputting the score of the growth condition of the mangrove forest, and generating a mangrove forest quality degradation area mask, a mangrove forest area reduction area mask and a change type map specifically includes: Performing channel splicing processing on the high-confidence-degree features and the mangrove evolution type probability distribution vector to obtain a spliced feature vector; feature fusion and information extraction are carried out on the spliced feature vectors through a multi-layer convolution network, so that fused semantic feature vectors are obtained; deep space refinement is carried out on the fused semantic feature vectors through a single branch scoring network, so that the mangrove growth condition score is obtained; and combining the mangrove growth condition score with the mangrove evolution type probability distribution vector, and carrying out identification treatment on the mangrove quality degradation area and the mangrove area reduction area to obtain a mangrove quality degradation area mask, a mangrove area reduction area mask and a change type map.
8. The method for monitoring the growth condition of the mangrove forest based on deep learning as set forth in claim 7, wherein the step of calculating the regional mangrove forest growth condition index and generating a mangrove forest hierarchical management suggestion based on the mangrove forest growth condition score and the mangrove forest evolution type probability distribution vector to realize the monitoring of the growth condition of the mangrove forest specifically includes: Determining an average growth condition score of a mangrove area according to a mangrove candidate growth area mask and the mangrove growth condition score, determining a mangrove healthy growth path occupation ratio according to a flourishing healthy type and an expanding new type response probability in a mangrove evolution type probability distribution vector, and determining a mangrove quality degradation and area reduction path occupation ratio according to a damaged degradation type and an atrophy breaking type response probability in the mangrove evolution type probability distribution vector; Constructing an regional mangrove growth condition index by combining the average mangrove growth condition score, the mangrove healthy growth path ratio, the mangrove quality degradation and the area reduction path ratio of the mangrove region; Based on the mangrove growth condition score and the mangrove evolution type probability distribution vector, a mangrove hierarchical management suggestion is generated, a mangrove evaluation result is output, and the mangrove growth condition is monitored, wherein the mangrove evaluation result comprises a mangrove growth condition score, a mangrove quality degradation area mask, a mangrove area reduction area mask, a change type map, an area mangrove growth condition index and the mangrove hierarchical management suggestion.
9. The method for monitoring the growth condition of the mangrove forest based on deep learning as set forth in claim 8, wherein the expression of the regional mangrove forest growth condition index is specifically as follows: In the above-mentioned method, the step of, Represents the regional mangrove growth status index, 、、 The weight coefficient is represented by a number of weight coefficients, Represents the average growth status score of the mangrove area, Representing the healthy growth path duty cycle of mangrove, Representing mangrove quality degradation and area reduction path occupancy.
10. Mangrove growth condition monitoring system based on deep learning is characterized by comprising the following modules: the first module is used for acquiring a first time-phase remote sensing image and a second time-phase remote sensing image of a target mangrove growth area, and extracting a mangrove candidate growth area and a double-time-phase depth semantic feature to obtain mangrove structural growth change features; the second module is used for carrying out feature matching on the structural growth change features of the mangrove based on a preset evolution mode feature set to obtain a mangrove evolution type probability distribution vector; the third module is used for constructing a ternary characteristic sequence and inputting the ternary characteristic sequence into the growth change rationality evaluation unit for rationality evaluation to obtain high-confidence-degree characteristics; and the fourth module is used for carrying out mangrove growth condition assessment and regional mangrove growth condition index calculation based on the high confidence coefficient characteristics and the mangrove evolution type probability distribution vector so as to realize the monitoring of the mangrove growth condition.

Description

Mangrove growth condition monitoring method and system based on deep learning Technical Field The invention relates to the technical field of remote sensing image processing, in particular to a mangrove growth condition monitoring method and system based on deep learning. Background Mangrove forest is used as a unique vegetation community growing in tropical and subtropical intertidal zones and has important ecological functions of wind prevention, wave elimination, carbon fixation, carbon storage, water quality purification, biodiversity maintenance and the like. However, the global mangrove ecosystem is facing a serious threat of degradation, subject to human activity disturbances (e.g., reclamation of the sea, expansion of the pond, environmental pollution) and climate changes (e.g., sea level rise, extreme weather). Developing high-precision monitoring and evaluation of the growth condition of the mangrove, generating a mangrove hierarchical management suggestion, and providing scientific basis for the establishment of ecological protection restoration and sustainable management policies of the mangrove. Traditional mangrove monitoring means rely mainly on field sampling surveys. Although the method is high in precision, the method is limited by objective factors such as poor accessibility of tidal flat terrain, high labor cost, long investigation period and the like, and quick and synchronous evaluation of mangrove growth conditions on a large space scale is difficult to realize. In recent years, with the development of remote sensing technology, remote sensing images have become an important data source for mangrove monitoring. Early researchers mostly use optical remote sensing images with single time phase to extract the spatial distribution range of mangrove by a supervised classification or non-supervised classification method, but the dynamic change process of mangrove along with time cannot be revealed. Partial researches apply a simple change detection method to compare mangrove distribution diagrams of two or more time phases, calculate the area increment and decrement, but only pay attention to the change of the mangrove range, and neglect the quality degradation phenomena of growth degradation, activity degradation and the like possibly occurring in the mangrove range in the unchanged area of the mangrove range. In addition, the scholars also use the vegetation index to evaluate the vegetation greenness and vigor, but the traditional vegetation index is easy to saturate in high biomass areas such as mangrove, difficult to accurately quantify complex growth conditions, and insensitive to early stress response. And the spectrum index designed manually is difficult to fully mine the deep semantic information of the remote sensing image, and has insufficient characterization capability on complex canopy structures and physiological changes of mangroves. Most researches are stopped at the current situation of monitoring or change drawing, and monitoring results cannot be effectively converted into layered management suggestions with space directivity and differentiation, so that a dislocation exists between technical achievements and actual business demands of ecological protection. Disclosure of Invention In order to solve the technical problems, the invention aims to provide a mangrove growth condition monitoring method and system based on deep learning, which can comprehensively identify the change of the mangrove range dynamic and internal growth condition and automatically generate mangrove hierarchical management suggestions. The first technical scheme adopted by the invention is that the mangrove growth condition monitoring method based on deep learning comprises the following steps: Acquiring a first time phase remote sensing image and a second time phase remote sensing image of a target mangrove growth area, and performing mangrove candidate growth area extraction processing and double time phase depth semantic feature extraction processing to obtain mangrove structural growth change features; Based on a preset evolution mode feature set, feature matching is carried out on the mangrove structured growth change features to obtain a mangrove evolution type probability distribution vector; constructing a ternary characteristic sequence and inputting the ternary characteristic sequence into a growth change rationality evaluation unit for rationality evaluation to obtain high-confidence characteristics; Based on the high confidence characteristic and the mangrove evolution type probability distribution vector, mangrove growth condition assessment and regional mangrove growth condition index calculation are carried out, and the mangrove growth condition is monitored. Further, the step of obtaining a first time-phase remote sensing image and a second time-phase remote sensing image of a target mangrove growth area, and performing mangrove candidate growth area extraction processing and double t