Port the SAR3D background gate, CLEAN and object boxes from Multi_depthnorm_clean
- BG_GATE_ENABLE: frame peak over its median background (sar3d_background); the GUI scales the frame with bg_gate_gain and its cloud threshold. - SAR_CLEAN_ENABLE: CLEAN of a one-frame aperture on a 4x10x4 cm search grid over the whole scene; objects in their own scale (sar3d_objects). - SAR_MARK_ENABLE: boxes by class around object groups (sar3d_marks), with the legend under the views. - SAR_VOXEL_M as three steps X/Y/Z with the notebook defaults 0.025/0.02/0.02. - GPU backend: image/download/load/project API, per-plane block reduction for the maximum, depth gate without FMA; sar3d host code built with -ffp-contract=off. - References regenerated from notebook cells 1 and 5, with new CLEAN scenes and a synthetic box-rule case; docs updated.
This commit is contained in:
Ayzen
@@ -71,13 +71,13 @@ controls, processing settings and a copy of the live plot through an embedded we
|
||||
- **gpr** — coherent back-projection on the ground plane, as in `Horns_motion_4x4_ground.py`,
|
||||
with object detection. Radar height and look angle define the slice in radar X/Z axes;
|
||||
the wave speed is 3×10⁸ m/s (air).
|
||||
- **sar3d** — 3D SAR of the reference notebook `Multi_antenna_demonstrate_depthnorm.ipynb`
|
||||
- **sar3d** — 3D SAR of the reference notebook `Multi_depthnorm_clean.ipynb`
|
||||
on an NVIDIA GPU through CUDA: front (X-Y), side (Z-Y), top (X-Z)
|
||||
and oblique maximum projections of the imaged
|
||||
volume. Its settings carry the notebook's parameter names; the scene window is
|
||||
set by the selected antennas with a 2 m X margin and height −0.5…2 m above the ground; its range extent follows
|
||||
`MIN_DEPTH`/`MAX_DEPTH`, and the voxels are cubes with edge `SAR_VOXEL_M`
|
||||
(default 2 cm). The radar motion is corrected from each trace's capture
|
||||
`MIN_DEPTH`/`MAX_DEPTH`, and the voxel steps `SAR_VOXEL_M` X, Y and Z default to the
|
||||
notebook's 2.5, 2 and 2 cm. The radar motion is corrected from each trace's capture
|
||||
timestamps, so the sweep timing needs no parameters.
|
||||
|
||||
Both coherent GPR and SAR apply `PAIR_DEPTH_EQUALIZE` (on by default) before the
|
||||
@@ -100,7 +100,27 @@ redraw the existing result without recalculating it. Tx/Rx markers appear on the
|
||||
front, side and top projections in ground coordinates. A hollow square marks the
|
||||
peak; its size follows the contiguous half-maximum width along the volume axes
|
||||
(FWHM, at least one voxel). This measures the imaged response rather than the
|
||||
reflector's physical dimensions.
|
||||
reflector's physical dimensions. The square is left out when CLEAN objects are shown
|
||||
or when the background gate empties the frame.
|
||||
|
||||
**BG_GATE_ENABLE** shows a frame relative to its background, the median of the volume
|
||||
over its working voxels: below 6 backgrounds the frame is empty, at 6 its maximum
|
||||
appears at the cloud threshold, and from 8.5 the scale is full. **SAR_CLEAN_ENABLE**
|
||||
runs the notebook's CLEAN on a one-frame aperture: always with `SAR_FRAME_RADIUS` 0,
|
||||
otherwise while no previous frame is available, as after Start. It subtracts the strongest point response
|
||||
from the A-scans of every pair, step by step on a 4 × 10 × 4 cm grid over the whole scene
|
||||
region, until the residual is background. The search ignores `SAR_X_SIZE_M` and
|
||||
`SAR_Z_SIZE_M`, so reflectors just outside a narrowed scene are subtracted too. Each object is then imaged apart and drawn in its own 0–1
|
||||
scale over the residual, so a weak object is not hidden by the sidelobes of a strong
|
||||
one. **SAR_MARK_ENABLE** draws boxes around groups of CLEAN objects: red for very low
|
||||
objects (like a mine), orange on the ground, blue above it, dashed grey at the scene
|
||||
boundary, where the response may lie partly outside. The legend is in the title of the
|
||||
3D view. Without boxes the objects are numbered with their level in backgrounds. All
|
||||
three switches are on by default, like in the notebook, and the status line reports the
|
||||
background ratio and the CLEAN result. The gate and CLEAN are processing settings; boxes
|
||||
only redraw. Their thresholds and rules are the notebook's fixed values. Both cost time:
|
||||
on a GTX 1060 with the default scene of 10 million voxels, the gate adds about 0.1 s per
|
||||
frame and CLEAN about 0.4 s more.
|
||||
|
||||
The desktop uses a light theme with white plot canvases. **Mode** stays below the
|
||||
**Actions** buttons while the settings sections scroll; the settings panel and
|
||||
@@ -222,7 +242,7 @@ The preprocessor still requires S21 calibration/reference sets. Start the GUI:
|
||||
Open **Preprocessing…**, use **Create Neutral S21 Sets**, then **Start**. This makes
|
||||
identity calibration + zero reference, preserving the simulated S21 samples.
|
||||
|
||||
Finer voxels increase SAR3D work as 1/`SAR_VOXEL_M`³. Increase the voxel size when
|
||||
SAR3D work grows inversely with the product of the three `SAR_VOXEL_M` steps. Increase them when
|
||||
a faster preview matters more than spatial resolution. Processing failures appear
|
||||
above the plots and a failed single capture
|
||||
finishes with its explanation while retaining the acquired data.
|
||||
@@ -469,19 +489,24 @@ physical timing. Real devices, GPIO and the CUDA backend need their own hardware
|
||||
|
||||
`check-sar3d` runs the real CUDA processor (`data_acq_and_processing/tests/test_backprojection.py`).
|
||||
It compares it with `sar3d_reference.npz`: compact float32/complex64 inputs, the three
|
||||
maximum projections of the notebook's volume, its joint peak/level and response
|
||||
widths for six compensation/normalization, equalization, background, scene and
|
||||
timing cases. The expected values come from unchanged functions in cells 1 and 14
|
||||
maximum projections of the notebook's displayed volume, its peak/level, response
|
||||
widths and background ratio for ten cases. Six cover compensation/normalization,
|
||||
equalization, background, scene and timing. Four are frames of point reflectors in noise
|
||||
for CLEAN, which also compare the objects and their boxes: with and without the background
|
||||
gate, with a scene where the stop at each range's background adds steps, and with a
|
||||
narrowed scene. The box rules are also checked
|
||||
one by one on synthetic objects. The expected values come from unchanged functions in cells 1 and 5
|
||||
of the original notebook, with each frame's previous-collection mean subtracted
|
||||
from its input spectra by NumPy. The same test checks the published
|
||||
status for invalid input, a result too large for the ring slot, an empty and a very weak
|
||||
signal. Without a usable GPU it checks only the CUDA warning and skips the rest; the
|
||||
notebook itself is not needed to run it. To regenerate the reference, pass the
|
||||
original notebook (it is kept outside the repository):
|
||||
status for invalid input, a volume beyond float32, a result too large for the ring slot,
|
||||
an empty and a very weak signal, and that the views are normalized at their peak on
|
||||
finer grids. Without a usable GPU it checks only the CUDA warning and the box rules and
|
||||
skips the rest; the notebook itself is not needed to run it. To regenerate the references,
|
||||
pass the original notebook and the Horns script (both are kept outside the repository):
|
||||
|
||||
```bash
|
||||
.venv/bin/python data_acq_and_processing/tests/generate_backprojection_reference.py \
|
||||
--notebook /path/to/Multi_antenna_demonstrate_depthnorm.ipynb \
|
||||
--notebook /path/to/Multi_depthnorm_clean.ipynb \
|
||||
--gpr-script /path/to/Horns_motion_4x4_ground.py
|
||||
```
|
||||
|
||||
|
||||
Reference in New Issue
Block a user