geoskill-hyperspectral-unmixing

VCA/N-FINDR端元提取+nnls线性解混丰度反演

高光谱端元提取与解混 | Hyperspectral Endmember Extraction & Unmixing

Extracts pure endmember spectra from hyperspectral imagery and inverts per-pixel abundances. Two classic endmember-extraction algorithms are supported:

• VCA (Vertex Component Analysis): iteratively seeks the projection direction with maximum variance in PCA-projected space and takes the extreme pixels as endmembers;
• N-FINDR (simplified): a greedy iterative search for the pixel combination that maximizes the simplex volume.

Linear unmixing uses FCLSU: for each pixel, scipy.optimize.nnls solves for non-negative least-squares abundances, which are then normalized to sum to 1 (abundance sum-to-one constraint); a residual map is also output to evaluate fit quality. Suitable for mixed-pixel decomposition, mineral/vegetation abundance mapping, and sub-pixel analysis.

Dependencies / 依赖

bash

pip install numpy rasterio scipy

Usage / 使用方法

Basic Usage (Synthetic Data, Offline)

bash

python geoskill-hyperspectral-unmixing.py --bbox 116.0 39.0 117.0 40.0 --synthetic --output-dir ./out

Example 1: VCA with 3 Endmembers

bash

python geoskill-hyperspectral-unmixing.py \
    --bbox 116.0 39.0 117.0 40.0 \
    --synthetic --n-endmembers 3 --method vca \
    --output-dir ./vca_3em

Example 2: N-FINDR Method Comparison

bash

python geoskill-hyperspectral-unmixing.py \
    --bbox 121.0 31.0 122.0 32.0 \
    --synthetic --n-endmembers 4 --method nfindr \
    --output-dir ./nfindr_4em

Example 3: More Bands

bash

python geoskill-hyperspectral-unmixing.py \
    --bbox 116.0 39.0 117.0 40.0 \
    --synthetic --n-bands 40 --n-endmembers 3 \
    --output-dir ./vca_40bands

Example 4: Real Hyperspectral GeoTIFF Input

bash

python geoskill-hyperspectral-unmixing.py \
    --input cuprite_subset.tif --n-endmembers 5 --method vca \
    --output-dir ./real_vca

Output / 输出

File	Format	Description
abundances.tif	GeoTIFF (float32, one band per endmember)	Normalized abundance map [0,1], EPSG:4326
residual.tif	GeoTIFF (float32)	Linear unmixing residual (RMSE) map
endmembers.json	JSON	Endmember spectra, algorithm, matching diagnostics
output-manifest.json	JSON	Run manifest (input/output/QA/software versions)

Data Source / 数据源 / Source

• Real mode: local hyperspectral GeoTIFF
• Synthetic mode: defines endmember spectra locally and generates pixels by linear mixing with random abundances plus noise; no external data source

Privacy / 隐私声明 / Privacy

• Runs fully offline by default and makes no network requests
• --synthetic mode reads no external data
• All computation is done locally; user data is never uploaded

License / License

MIT

---

---

name: geoskill-hyperspectral-unmixing description: 'VCA/N-FINDR端元提取+nnls线性解混丰度反演'

高光谱端元提取与解混 | Hyperspectral Endmember Extraction & Unmixing

从高光谱影像中提取纯净端元（endmember）光谱并反演逐像元丰度 （abundance）。端元提取支持两种经典算法：

• VCA（Vertex Component Analysis）：在 PCA 投影空间中迭代寻找 方差最大的投影方向，取极值像元作为端元；
• N-FINDR（简化版）：贪心迭代搜索使单纯体体积最大的像元组合。

线性解混采用 FCLSU：对每个像元用 scipy.optimize.nnls 求非负最小 二乘丰度，再归一化到和为 1（丰度总和约束），并输出残差图评估拟合 质量。适用于混合像元分解、矿物/植被丰度制图、亚像元分析。

依赖

bash

pip install numpy rasterio scipy

使用方法

基本用法（合成数据，离线）

bash

python geoskill-hyperspectral-unmixing.py --bbox 116.0 39.0 117.0 40.0 --synthetic --output-dir ./out

示例 1：VCA 提取 3 个端元

bash

python geoskill-hyperspectral-unmixing.py \
    --bbox 116.0 39.0 117.0 40.0 \
    --synthetic --n-endmembers 3 --method vca \
    --output-dir ./vca_3em

示例 2：N-FINDR 方法对比

bash

python geoskill-hyperspectral-unmixing.py \
    --bbox 121.0 31.0 122.0 32.0 \
    --synthetic --n-endmembers 4 --method nfindr \
    --output-dir ./nfindr_4em

示例 3：更多波段

bash

python geoskill-hyperspectral-unmixing.py \
    --bbox 116.0 39.0 117.0 40.0 \
    --synthetic --n-bands 40 --n-endmembers 3 \
    --output-dir ./vca_40bands

示例 4：真实高光谱 GeoTIFF 输入

bash

python geoskill-hyperspectral-unmixing.py \
    --input cuprite_subset.tif --n-endmembers 5 --method vca \
    --output-dir ./real_vca

输出

文件	格式	说明
abundances.tif	GeoTIFF (float32, 每端元一波段)	归一化丰度图 [0,1]，EPSG:4326
residual.tif	GeoTIFF (float32)	线性解混残差（RMSE）图
endmembers.json	JSON	端元光谱、算法、匹配诊断
output-manifest.json	JSON	运行清单（输入/输出/QA/软件版本）

数据源 / Source

• 真实模式：本地高光谱 GeoTIFF
• 合成模式：本地定义端元光谱，按随机丰度线性混合 + 噪声生成像元， 无外部数据源

隐私声明 / Privacy

• 默认完全离线运行，不发起任何网络请求
• --synthetic 模式不读取任何外部数据
• 所有计算在本地完成，不上传用户数据

License

MIT