请留言
slide1

Global Vision Media Focus

The company was featured on CCTV a total of 15 times, and its reach through Party media, central media, and local official media reached a total of 50 million people.

0101
News Categories

The "Revelation" of Spartina alterniflora: A Three-Step Process from Coarse Extraction to Refined Identification

2026-07-31

Along Fujian's long coastline, an invasive plant is quietly changing the face of wetlands—Spartina alterniflora. With its extensive root system and rapid expansion, it encroaches on the habitat of native vegetation such as reeds and mangroves, impacting bird habitats and wetland ecological functions. Accurately understanding its distribution is the first step towards scientific management.

Based on the ZY1E hyperspectral data from June 2020, we gradually incorporated Sentinel-2 optical temporal features and Sentinel-1 radar information, completing a full technical upgrade from "coarse extraction" to "refined identification." The entire process can be clearly divided into three steps, with accuracy and spatial integrity improving step by step.

Hyperspectral Random Forest Classification

We first used ZY1E hyperspectral imagery (June 17, 2020), combined with ground training samples (Spartina alterniflora, bare land, other vegetation, buildings, farmland, etc.), to run a random forest algorithm to obtain multi-class classification results. This step can quickly identify...Spartina alternifloraThe main distribution area of ​​Spartina alterniflora was identified, but due to the similarity of spectral characteristics between Spartina alterniflora and local salt marsh plants, a significant amount of other wetland vegetation was mixed in with the results, and the patches were relatively fragmented. This is a typical "rough extraction" stage.

Figure 1.ZY1EPreliminary classification of hyperspectral random forest(Orange represents Spartina alterniflora)

Phenological rules refinement

Local research in Fujian shows that June and July are the months with the greatest spectral differences between Spartina alterniflora and other wetland vegetation. Our ZY1E imagery falls precisely within this window.

So weGIVEThe study incorporates Sentinel-2 long-term series data, focusing on extracting vegetation indices (NDVI, EVI, red edge index, etc.) from June to July, and also includes winter greenness information. The core logic is:

• Spartina alterniflora shows a significant increase in greenness at the beginning of the growing season;

• Mangroves wither completely in winter, and their greenness is significantly lower than that of evergreen mangroves;

• Red bandIt is more sensitive to chlorophyll and leaf structure, which helps to distinguish vegetation with similar morphology.

By using these phenological rules to perform a secondary filtering on the results of the first step, a large number of misclassified wetland vegetation were effectively eliminated, and the recognition accuracy was significantly improved.

Figure 2.Results of Sentinel-2 phenological rule refinement(Red represents Spartina alterniflora)

Radar and SNIC post-processing

Even after filtering by phenological rules, the results still have certain limitations.Salt and pepper noiseAnd boundary burrs. We further introduced Sentinel-1 radar data (VV, VH backscattering and their ratios). Spartina alterniflora plants are taller and denser, and their radar echo characteristics differ from those of reeds and bare beaches, providing additional distinguishing information.

Simultaneously, SNIC superpixel segmentation is employed to group similar pixels together, and the determination is made by integrating optical phenological features and radar features at the object scale. The improvements brought about by this step are mainly reflected in:

• The number of fine, misclassified plaques has been further reduced;

• The boundaries of the patches are more complete and closely resemble actual ground features;

• The overall map is cleaner and more suitable for subsequent area statistics and remediation applications.

Figure 3.Sentinel-1 + SNIC Results(Red represents Spartina alterniflora)

Results and Application Value

After the above three progressive processing steps, we obtained a refined distribution map of Spartina alterniflora in some coastal areas of Fujian in 2020. From the original random forest result, to the refinement of phenological rules, and then to radar and object-oriented post-processing, the recognition accuracy gradually improved, the patch integrity continued to improve, and the final result is closer to the true distribution pattern.

Comparing the three result images clearly shows the changes brought about by the technological upgrade:

• Figure 1 (original classification) shows fragmented patches with significant confusion;

• Figure 2 (Phenological Refinement) shows a significant reduction in misclassification;

• Figure 3 (Final Result) shows more complete boundaries and a cleaner image.

These results can directly serve as a basis for surveying and monitoring the substrate of Spartina alterniflora, assessing the progress of cleanup and remediation projects, and making spatial decisions for wetland protection and ecological restoration. It is important to emphasize that remote sensing extraction still needs to be combined with field investigations, and local sample calibration remains crucial.

Conclusion

From the initial classification of a single hyperspectral image, to the introduction of optical phenological rules, and then to the fusion of radar information and...Object-oriented processingWe have completed the entire technical upgrade path for remote sensing extraction of Spartina alterniflora. Each step improved upon the shortcomings of the previous one, resulting in enhanced accuracy and spatial quality.

Technology itself is not the goal; its true value lies in providing more reliable spatial information for wetland protection and management. Controlling Spartina alterniflora is a long-term endeavor; only with a precise understanding of the wetland's condition can subsequent prevention, eradication, and ecological restoration be more targeted and effective.