Upload source code snapshot

README.md

@@ -0,0 +1,92 @@
+# Typical Marine Ecological Environment Feature Recognition
+
+典型海洋生态环境地物要素识别项目。项目目标不是单一浒苔二分类，而是把多种海洋生态环境要素统一到一个可训练、可推理、可扩展的遥感识别框架中。
+
+当前参考/整合方向来自 `BiJiaNet` 中的几个任务：
+
+- `EntGreenTide`: 浒苔/绿潮。
+- `RedTide`: 赤潮。
+- `SargGoldenTide`: 马尾藻/金潮。
+- `AquacultureArea`: 养殖区、养殖设施或相关海上人工目标。
+
+## Core Constraint
+
+不假设有海岸线、陆地掩膜、外部矢量层或其他 GIS 先验可用。
+
+陆地、黑区、无效影像区、云雾、水体背景等必须通过数据标签、hard negative 样本、影像有效性规则和模型上下文学习来处理，不能依赖外部矢量后处理。
+
+## Target Classes
+
+推荐统一语义类别：
+
+| ID | Class |
+| --- | --- |
+| 0 | other/background |
+| 1 | invalid/no-data/black area |
+| 2 | water |
+| 3 | land/non-water hard negative |
+| 4 | cloud/haze/shadow |
+| 5 | green_tide |
+| 6 | red_tide |
+| 7 | golden_tide |
+| 8 | aquaculture |
+
+旧二分类标签不会被直接解释成“所有其他类别都为负样本”。导入时需要保留其来源、要素类型和标签语义。
+
+## Data Strategy
+
+数据集采用 manifest 驱动，允许并保留：
+
+- 128、256、512 等不同 tile 尺度。
+- GF1、GF2、GF6 等不同卫星。
+- PMS、MUX、MSS、PAN 等不同传感器。
+- 2 米、8 米等不同空间分辨率。
+- 已融合影像和 PAN+MSS 原始组合。
+- 全景影像、切片影像、二值 mask、多类别 mask、推理结果。
+
+详细规范见 [docs/dataset_standard.md](docs/dataset_standard.md)。
+
+## Key Scripts
+
+- `scripts/build_marine_feature_dataset.py`: 扫描服务器或本地数据根目录，生成规范化 `assets_raw.jsonl`、`samples.jsonl` 和 `asset_inventory.csv`。
+- `scripts/infer_whole_scene.py`: 整景滑窗推理，支持重叠加权、条带写出、可选 PAN+MSS 边融合边推理。
+- `scripts/postprocess_mask.py`: 临时后处理脚本，只作为黑区/明显非目标区域清理工具，不能替代最终多类别模型。
+- `train_seaweed_segmentation.py`: 当前遗留 DINOv3 + DeepLabV3+ 训练入口，后续会重构为多要素多头训练入口。
+- `dinov3_deeplabv3plus.py`: 当前 DINOv3 编码器 + DeepLabV3+ 解码器实现。
+
+## Normalized Dataset Example
+
+```powershell
+python scripts/build_marine_feature_dataset.py `
+  --roots D:\hutai-2 D:\Dataset E:\Dataset `
+  --output-root D:\marine_feature_dataset
+```
+
+输出：
+
+```text
+D:\marine_feature_dataset\
+  manifests\
+    assets_raw.jsonl
+    samples.jsonl
+  reports\
+    asset_inventory.csv
+```
+
+## Near-Term Refactor Plan
+
+1. 盘点服务器上所有可用要素数据，生成数据资产清单。
+2. 统一 manifest 格式，保留尺度、卫星、传感器、分辨率、融合状态等元数据。
+3. 将当前浒苔二分类模型升级为多类别/多头模型：
+   - valid/invalid head
+   - scene context head: water, land, cloud/shadow, other
+   - ecological feature head: green tide, red tide, golden tide, aquaculture
+4. 改造训练 sampler，使不同尺度、不同传感器数据能按元数据混合训练。
+5. 改造整景推理输出，多通道概率图和最终要素 mask 分开保存。
+
+## Notes
+
+- `data/`、`outputs/`、大影像和模型 checkpoint 不进入 git。
+- GF2 这类 PAN+MSS 数据应在推理时边融合边推理，避免落盘整幅融合图。
+- GF6 这类已融合图可以直接整景滑窗推理。
+- 当前 GF6 浒苔结果只能作为候选区，不能作为最终产品，因为旧模型没有显式学习陆地、黑区和其他生态要素类别。

configs/marine_feature_multitask.json

@@ -0,0 +1,48 @@
+{
+  "project_name": "Typical Marine Ecological Environment Feature Recognition",
+  "task": "marine_feature_multitask_segmentation",
+  "num_classes": 9,
+  "classes": [
+    "other",
+    "invalid",
+    "water",
+    "land",
+    "cloud_shadow",
+    "green_tide",
+    "red_tide",
+    "golden_tide",
+    "aquaculture"
+  ],
+  "input": {
+    "supported_patch_sizes": [128, 256, 512, 1024],
+    "default_train_size": 256,
+    "supported_satellites": ["GF1", "GF2", "GF6", "HY", "Sentinel2", "Landsat"],
+    "supported_sensors": ["PMS", "MUX", "MSS", "PAN", "WFV"],
+    "default_bands": ["blue", "green", "red", "nir"],
+    "allow_pan_mss_streaming_fusion": true,
+    "allow_fused_products": true
+  },
+  "model": {
+    "encoder": "dinov3_vitl16",
+    "decoder": "deeplabv3plus",
+    "heads": {
+      "semantic": 9,
+      "validity": 2,
+      "context": 5,
+      "ecological_element": 4
+    }
+  },
+  "loss": {
+    "semantic_weight": 1.0,
+    "validity_weight": 0.3,
+    "context_weight": 0.5,
+    "element_weight": 1.0,
+    "ignore_index": 255
+  },
+  "dataset": {
+    "manifest": "data_marine_features/manifests/samples.jsonl",
+    "do_not_assume_external_land_or_coastline_mask": true,
+    "preserve_native_resolution_metadata": true,
+    "preserve_native_patch_size_metadata": true
+  }
+}

docs/dataset_standard.md

@@ -0,0 +1,158 @@
+# Typical Marine Ecological Environment Feature Dataset Standard
+
+This project uses a manifest-driven dataset layout for marine ecological environment
+feature recognition. It is designed to keep different patch sizes, satellites,
+sensors, spatial resolutions, and fusion states instead of forcing everything into
+one fixed tile format.
+
+## Scope
+
+Target feature families:
+
+- `green_tide`: Enteromorpha / green tide / seaweed.
+- `red_tide`: red tide / harmful algal bloom.
+- `golden_tide`: Sargassum / golden tide.
+- `aquaculture`: aquaculture area, rafts, cages, ponds, or related facilities.
+- `water`: valid water background.
+- `land`: land or non-water hard negative.
+- `invalid`: black border, no-data, saturated, missing, or otherwise unusable pixels.
+- `cloud_shadow`: cloud, haze, cloud shadow, or bright atmospheric interference.
+- `other`: valid but not assigned to the categories above.
+
+No external coastline, land-mask vector, or GIS mask is assumed to exist. Land,
+invalid area, and water/background handling must be represented by dataset labels,
+hard-negative samples, or image-derived validity rules.
+
+## Canonical Directory Layout
+
+```text
+data_marine_features/
+  manifests/
+    assets_raw.jsonl
+    samples.jsonl
+    splits/
+      train.txt
+      val.txt
+      test.txt
+  images/
+    <asset_id>.<ext>          # optional symlink or copied tile
+  masks/
+    <sample_id>.<ext>         # optional symlink or copied mask
+  previews/
+    <sample_id>.png
+  reports/
+    asset_inventory.csv
+```
+
+The canonical source of truth is `manifests/samples.jsonl`. Files may remain in
+their original locations; `images/` and `masks/` are optional conveniences.
+
+## `samples.jsonl` Schema
+
+Each line is one image sample or tile.
+
+```json
+{
+  "sample_id": "gf6_20250604_green_tide_000001",
+  "element": "green_tide",
+  "task_type": "semantic_segmentation",
+  "image_path": "D:/.../image.tif",
+  "mask_path": "D:/.../mask.tif",
+  "label_encoding": {"0": "background", "1": "green_tide"},
+  "satellite": "GF6",
+  "sensor": "PMS",
+  "resolution_m": 2.0,
+  "patch_size": 256,
+  "bands": ["blue", "green", "red", "nir"],
+  "band_count": 4,
+  "dtype": "uint16",
+  "is_fused": true,
+  "has_pan": false,
+  "pan_path": null,
+  "acquired_at": "2025-06-04",
+  "source_project": "EntGreenTide",
+  "source_dataset": "GF6_PMS_E121.1_N33.6_20250604_L1A1420584616",
+  "split": "train",
+  "quality_flags": ["valid_image"],
+  "notes": ""
+}
+```
+
+Required fields:
+
+- `sample_id`
+- `element`
+- `task_type`
+- `image_path`
+- `satellite`
+- `sensor`
+- `patch_size`
+- `band_count`
+- `is_fused`
+- `source_project`
+
+Recommended fields:
+
+- `mask_path`
+- `resolution_m`
+- `bands`
+- `dtype`
+- `has_pan`
+- `pan_path`
+- `acquired_at`
+- `split`
+- `quality_flags`
+
+## Multi-Scale Policy
+
+Patch sizes such as 128, 256, 512, and full-scene windows are all valid. They are
+not merged destructively. Training code should sample them with metadata-aware
+transforms:
+
+- Keep `patch_size` in the manifest.
+- Resize only inside the training transform when the model requires it.
+- Preserve the original spatial resolution in `resolution_m`.
+- For full-scene inference, use sliding windows whose size is a runtime parameter.
+
+## Multi-Sensor Policy
+
+Different satellites and sensors are expected:
+
+- GF1, GF2, GF6, and other optical satellites can coexist.
+- Fused PMS/MUX products and raw PAN+MSS products can coexist.
+- If PAN+MSS are available, keep both paths and mark `has_pan=true`.
+- If already fused, mark `is_fused=true`.
+- Do not assume band order from the filename alone; record `bands` when known.
+
+## Label Policy
+
+The final model should not be binary seaweed/background. It should learn ecological
+elements and scene context. A recommended unified semantic target is:
+
+| Class ID | Name |
+| --- | --- |
+| 0 | other/background |
+| 1 | invalid |
+| 2 | water |
+| 3 | land |
+| 4 | cloud_shadow |
+| 5 | green_tide |
+| 6 | red_tide |
+| 7 | golden_tide |
+| 8 | aquaculture |
+
+When legacy binary masks are imported, they should be represented as task-specific
+labels plus metadata. Do not silently treat unlabeled pixels as true negatives for
+all other ecological elements.
+
+## Inventory Rules
+
+The inventory scanner should:
+
+- Preserve original file paths.
+- Infer element type from directory names and known project folders.
+- Infer satellite/sensor/acquisition date from filenames when possible.
+- Pair image and mask files by stem when masks exist.
+- Record unpaired full-scene images as inference assets.
+- Flag ambiguous data instead of guessing labels.
+

typical-marine-ecological-feature-recognition-source.zip

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9533c76425b89d40710a2268484d6a9a93683566eb8d465ab7a42307275f1d4c
+size 404413