stage02: filtre strict v5 (pct=80 dur=60 depth=-3m) + stage02b diag plots

Defaults plus stricts pour éliminer surface/yoyo:
- min_near_bottom_pct: 50 -> 80 %
- min_sustained_duration: 30 -> 60 s
- min_mission_depth: -2 -> -3 m
- min_displacement_m: 5.0 documenté (stage06+ futur)

Nouveau script 02b_runs_diag.py: 4-panel PNG par run OK+rejected
(rel_alt+threshold, MAVROS state, depth histo, verdict criteria)
+ index.html pour inspection visuelle Flag.

Test Lepradet: 5 -> 1 run OK (AUV210_run_00 79s -13m 81pct)
Page publiée: laboratoire.freeboxos.fr/02-runs-diag-lepradet/

pipeline/stages/02_mission_run_detect.py

@@ -47,9 +47,10 @@ DEFAULT_STATE_MODES = {"AUTO", "GUIDED"}
 STATE_STOP_MODES = {"SURFACE", "MANUAL"}
 
 # Filtrage "vraie mission" (modifiables via argparse)
-DEFAULT_MIN_MISSION_DEPTH = -2.0   # rel_alt doit atteindre ce seuil (m)
+DEFAULT_MIN_MISSION_DEPTH = -3.0   # rel_alt doit atteindre ce seuil (m) — strict v5 2026-05-14
-DEFAULT_MIN_SUSTAINED_DURATION = 30.0  # pendant au moins N secondes consécutives
+DEFAULT_MIN_SUSTAINED_DURATION = 60.0  # pendant au moins N secondes consécutives — strict v5
-DEFAULT_MIN_NEAR_BOTTOM_PCT = 0.0  # % min du run où rel_alt < min_mission_depth (0 = désactivé)
+DEFAULT_MIN_NEAR_BOTTOM_PCT = 80.0  # % min du run où rel_alt < min_mission_depth — strict v5
+DEFAULT_MIN_DISPLACEMENT_M = 5.0   # futur stage06+: déplacement min USBL — documenté seulement v5
 
 # Filtrage "avant première plongée"
 DEFAULT_FIRST_SUBMERSION_DEPTH = -2.0   # seuil rel_alt pour considérer submergé

pipeline/stages/02b_runs_diag.py

@@ -0,0 +1,352 @@
+#!/usr/bin/env python3
+"""Stage 02b — Diagnostic plots per run candidate.
+
+Reads 02_runs.json + replays MCAP rel_alt + /mavros/state to produce
+4-panel PNG per run (validated + rejected).
+
+Output:
+    data/<MISSION>/02_runs_diag/<RUN_ID>.png
+    data/<MISSION>/02_runs_diag/index.json
+    data/<MISSION>/02_runs_diag/index.html
+
+Usage:
+    python3 02b_runs_diag.py --mission 20260505-Lepradet \
+        [--ssd-base /mnt/ssd] [--out data/] [--padding-s 120]
+"""
+
+import argparse
+import json
+import re
+import sys
+from datetime import datetime, timezone
+from pathlib import Path
+
+import matplotlib
+matplotlib.use("Agg")
+import matplotlib.pyplot as plt
+from matplotlib.patches import Rectangle
+
+from mcap.reader import make_reader
+from mcap_ros2.decoder import DecoderFactory
+
+TOPIC_REL_ALT = "/mavros/global_position/rel_alt"
+TOPIC_STATE = "/mavros/state"
+TOPIC_HEADING = "/mavros/global_position/compass_hdg"
+
+AUV_PHYSICAL_MAP = {"AUV010": "AUV210", "AUV012": "AUV212", "AUV013": "AUV213"}
+PHYS_TO_MCAP = {v: k for k, v in AUV_PHYSICAL_MAP.items()}
+
+
+def extract_auv_id(folder_name):
+    m = re.search(r"(AUV\d+)$", folder_name)
+    return m.group(1) if m else None
+
+
+def gather_signals_for_auv(bags_root, auv_mcap_id, t_start, t_end):
+    """Read rel_alt + state + heading from all bags for this AUV in [t_start,t_end]."""
+    alt = []  # (t, v)
+    state = []  # (t, armed, mode)
+    hdg = []  # (t, v_deg)
+    for d in sorted(bags_root.iterdir()):
+        if not d.is_dir():
+            continue
+        if extract_auv_id(d.name) != auv_mcap_id:
+            continue
+        for mcap_path in sorted(d.glob("*.mcap")):
+            try:
+                with open(mcap_path, "rb") as fp:
+                    reader = make_reader(fp, decoder_factories=[DecoderFactory()])
+                    for _schema, channel, message, ros_msg in reader.iter_decoded_messages(
+                            topics=[TOPIC_REL_ALT, TOPIC_STATE, TOPIC_HEADING]):
+                        ts = message.log_time / 1e9
+                        if ts < t_start or ts > t_end:
+                            continue
+                        if channel.topic == TOPIC_REL_ALT:
+                            alt.append((ts, float(ros_msg.data)))
+                        elif channel.topic == TOPIC_STATE:
+                            state.append((ts, bool(ros_msg.armed), str(ros_msg.mode)))
+                        elif channel.topic == TOPIC_HEADING:
+                            hdg.append((ts, float(ros_msg.data)))
+            except Exception as exc:
+                print(f"  [WARN] skip {mcap_path.name}: {exc}", file=sys.stderr)
+    alt.sort()
+    state.sort()
+    hdg.sort()
+    return alt, state, hdg
+
+
+def plot_run_diag(run, alt, state, hdg, filter_params, status, reject_reason, out_path):
+    """4-panel diagnostic plot."""
+    start_e = run["start_epoch"]
+    end_e = run["end_epoch"]
+    dur = end_e - start_e
+    run_id = run["run_id"]
+
+    # Convert epoch → relative seconds from start_e for x-axis readability
+    def to_rel(t):
+        return t - start_e
+
+    fig, axes = plt.subplots(4, 1, figsize=(12, 11), gridspec_kw={"height_ratios": [3, 1, 2, 2]})
+    fig.suptitle(
+        f"{run_id} — {status}  |  start={datetime.fromtimestamp(start_e, tz=timezone.utc).strftime('%H:%M:%S UTC')}  duration={dur:.0f}s",
+        fontsize=12, fontweight="bold",
+        color=("#16a34a" if status == "OK" else "#dc2626"),
+    )
+
+    # === Panel A: rel_alt time series ===
+    ax = axes[0]
+    if alt:
+        t_a = [to_rel(t) for t, _ in alt]
+        v_a = [v for _, v in alt]
+        ax.plot(t_a, v_a, color="#1e40af", lw=1.0, label="rel_alt")
+    threshold = filter_params["min_mission_depth_m"]
+    ax.axhline(threshold, color="#dc2626", ls="--", lw=1.0,
+               label=f"threshold {threshold}m")
+    ax.axhline(0, color="#94a3b8", ls=":", lw=0.8, label="surface")
+    # Mark run window
+    ax.axvspan(0, dur, alpha=0.10, color="#16a34a" if status == "OK" else "#dc2626")
+    ax.set_ylabel("rel_alt (m)")
+    ax.set_title(
+        f"(a) Depth — max={run['max_depth_m']}m  mean={run['mean_depth_m']}m  "
+        f"sustained={run['sustained_duration_s']:.0f}s below {threshold}m  "
+        f"pct_near_bottom={run['pct_near_bottom']:.1f}%"
+    )
+    ax.legend(loc="lower right", fontsize=8)
+    ax.grid(True, alpha=0.3)
+    ax.set_xlim(-30, dur + 30)
+
+    # === Panel B: state armed + mode ===
+    ax = axes[1]
+    if state:
+        t_s = [to_rel(t) for t, _, _ in state]
+        armed_y = [1 if a else 0 for _, a, _ in state]
+        ax.step(t_s, armed_y, where="post", color="#16a34a", lw=1.2, label="armed")
+        # Annotate dominant mode + mode transitions
+        prev_mode = None
+        for t, _, m in state:
+            if m != prev_mode:
+                ax.axvline(to_rel(t), color="#7c3aed", lw=0.5, alpha=0.4)
+                ax.text(to_rel(t), 0.5, m, rotation=90, fontsize=7,
+                        color="#7c3aed", va="center", alpha=0.7)
+                prev_mode = m
+    ax.set_ylim(-0.2, 1.3)
+    ax.set_yticks([0, 1])
+    ax.set_yticklabels(["disarmed", "armed"])
+    ax.set_title(f"(b) MAVROS state — dominant mode: {run.get('dominant_mode','?')}")
+    ax.grid(True, alpha=0.3)
+    ax.set_xlim(-30, dur + 30)
+
+    # === Panel C: depth distribution + cumulative time below threshold ===
+    ax = axes[2]
+    # filter alt to run window
+    run_vals = [v for t, v in alt if start_e <= t <= end_e]
+    if run_vals:
+        ax.hist(run_vals, bins=40, color="#3b82f6", alpha=0.7, edgecolor="white")
+        ax.axvline(threshold, color="#dc2626", ls="--", lw=1.0,
+                   label=f"threshold {threshold}m")
+        n_total = len(run_vals)
+        n_below = sum(1 for v in run_vals if v < threshold)
+        ax.set_title(
+            f"(c) Depth distribution within run  |  {n_below}/{n_total} samples below threshold "
+            f"= {100*n_below/n_total:.1f}%"
+        )
+        ax.legend(loc="upper right", fontsize=8)
+    ax.set_xlabel("rel_alt (m)")
+    ax.set_ylabel("samples")
+    ax.grid(True, alpha=0.3)
+
+    # === Panel D: verdict box ===
+    ax = axes[3]
+    ax.axis("off")
+    color = "#16a34a" if status == "OK" else "#dc2626"
+    verdict_text = f"VERDICT: {status}"
+    ax.text(0.02, 0.85, verdict_text, fontsize=18, fontweight="bold",
+            color=color, transform=ax.transAxes)
+
+    # Build criteria summary
+    crit_lines = []
+    min_dur = filter_params["min_sustained_duration_s"]
+    min_pct = filter_params["min_near_bottom_pct"]
+    min_depth = filter_params["min_mission_depth_m"]
+
+    pass_dur = run["sustained_duration_s"] >= min_dur
+    pass_pct = run["pct_near_bottom"] >= min_pct
+    pass_depth = run["max_depth_m"] < min_depth
+
+    crit_lines.append(
+        f"{'OK' if pass_depth else 'KO':3s}  max_depth {run['max_depth_m']}m < {min_depth}m"
+    )
+    crit_lines.append(
+        f"{'OK' if pass_dur else 'KO':3s}  sustained_duration {run['sustained_duration_s']:.0f}s >= {min_dur:.0f}s"
+    )
+    crit_lines.append(
+        f"{'OK' if pass_pct else 'KO':3s}  pct_near_bottom {run['pct_near_bottom']:.1f}% >= {min_pct:.0f}%"
+    )
+    crit_lines.append(
+        f"     duration {run['duration_s']:.0f}s | mode {run.get('dominant_mode','?')} | method {run.get('detection_method','?')}"
+    )
+
+    for i, line in enumerate(crit_lines):
+        ax.text(0.02, 0.65 - i * 0.13, line, fontsize=10, family="monospace",
+                transform=ax.transAxes)
+
+    if reject_reason:
+        ax.text(0.02, 0.05, f"REASON: {reject_reason}", fontsize=9,
+                color="#dc2626", transform=ax.transAxes, family="monospace")
+
+    axes[-2].set_xlabel("time since run start (s)")
+    plt.tight_layout()
+    plt.savefig(out_path, dpi=90, bbox_inches="tight")
+    plt.close(fig)
+
+
+HTML_TEMPLATE = """<!doctype html>
+<html lang="fr">
+<head>
+<meta charset="utf-8">
+<title>02 Runs Diagnostic — {mission}</title>
+<style>
+body {{ font-family: -apple-system, system-ui, sans-serif; background: #0f172a; color: #e2e8f0; margin: 0; padding: 20px; }}
+h1 {{ color: #38bdf8; margin: 0 0 4px 0; }}
+.subtitle {{ color: #94a3b8; margin-bottom: 20px; font-size: 14px; }}
+.params {{ background: #1e293b; padding: 10px 14px; border-radius: 6px; margin-bottom: 24px; font-family: monospace; font-size: 12px; color: #cbd5e1; }}
+.grid {{ display: grid; grid-template-columns: repeat(auto-fit, minmax(560px, 1fr)); gap: 20px; }}
+.card {{ background: #1e293b; border-radius: 8px; padding: 12px; border-left: 6px solid #475569; }}
+.card.ok {{ border-left-color: #16a34a; }}
+.card.ko {{ border-left-color: #dc2626; }}
+.card h3 {{ margin: 0 0 6px 0; font-size: 16px; }}
+.card h3.ok {{ color: #4ade80; }}
+.card h3.ko {{ color: #f87171; }}
+.card img {{ width: 100%; height: auto; border-radius: 4px; }}
+.meta {{ color: #94a3b8; font-size: 12px; margin: 4px 0 8px 0; font-family: monospace; }}
+.reason {{ color: #fca5a5; font-size: 12px; margin-top: 6px; font-family: monospace; }}
+.summary {{ display: flex; gap: 16px; margin-bottom: 20px; }}
+.pill {{ background: #1e293b; padding: 8px 14px; border-radius: 999px; font-size: 13px; }}
+.pill.ok {{ color: #4ade80; }}
+.pill.ko {{ color: #f87171; }}
+</style>
+</head>
+<body>
+<h1>02 Runs Diagnostic — {mission}</h1>
+<div class="subtitle">Generated {generated_at} · filter v5 strict</div>
+<div class="summary">
+  <div class="pill ok">OK: {n_ok}</div>
+  <div class="pill ko">Rejected: {n_ko}</div>
+  <div class="pill">Total candidates: {n_total}</div>
+</div>
+<div class="params">{params}</div>
+<div class="grid">
+{cards}
+</div>
+</body>
+</html>
+"""
+
+CARD_TEMPLATE = """<div class="card {cls}">
+  <h3 class="{cls}">{run_id} — {status}</h3>
+  <div class="meta">duration={duration}s | max_depth={max_depth}m | sustained={sustained}s | pct_near={pct}% | mode={mode}</div>
+  <img src="{png}" alt="{run_id}">
+  {reason_block}
+</div>"""
+
+
+def build_html(mission, filter_params, generated_at, ok_runs, ko_runs, out_dir):
+    cards = []
+    for run, status, reason in [(r, "OK", None) for r in ok_runs] + [(r, "REJECTED", r.get("rejected_reason")) for r in ko_runs]:
+        cls = "ok" if status == "OK" else "ko"
+        reason_block = f'<div class="reason">REASON: {reason}</div>' if reason else ""
+        cards.append(CARD_TEMPLATE.format(
+            cls=cls, run_id=run["run_id"], status=status,
+            duration=f"{run['duration_s']:.0f}",
+            max_depth=run["max_depth_m"],
+            sustained=f"{run['sustained_duration_s']:.0f}",
+            pct=f"{run['pct_near_bottom']:.1f}",
+            mode=run.get("dominant_mode", "?"),
+            png=f"{run['run_id']}.png",
+            reason_block=reason_block,
+        ))
+    html = HTML_TEMPLATE.format(
+        mission=mission,
+        generated_at=generated_at,
+        params=json.dumps(filter_params, indent=2),
+        n_ok=len(ok_runs),
+        n_ko=len(ko_runs),
+        n_total=len(ok_runs) + len(ko_runs),
+        cards="\n".join(cards),
+    )
+    (out_dir / "index.html").write_text(html, encoding="utf-8")
+
+
+def main():
+    parser = argparse.ArgumentParser(description="Stage 02b — runs diagnostic plots")
+    parser.add_argument("--mission", required=True)
+    parser.add_argument("--ssd-base", default="/mnt/ssd")
+    parser.add_argument("--out", default="data/")
+    parser.add_argument("--padding-s", type=float, default=120.0,
+                        help="padding (s) avant/après run pour contexte")
+    args = parser.parse_args()
+
+    mission_dir = Path(args.out) / args.mission
+    runs_json = mission_dir / "02_runs.json"
+    if not runs_json.exists():
+        print(f"[ERROR] {runs_json} missing — run 02_mission_run_detect first", file=sys.stderr)
+        sys.exit(1)
+
+    bags_root = Path(args.ssd_base) / args.mission / "raw_data" / "logs" / "SUB" / "bag"
+    if not bags_root.exists():
+        print(f"[ERROR] bags dir not found: {bags_root}", file=sys.stderr)
+        sys.exit(1)
+
+    data = json.loads(runs_json.read_text())
+    filter_params = data["filter_params"]
+    out_dir = mission_dir / "02_runs_diag"
+    out_dir.mkdir(parents=True, exist_ok=True)
+
+    ok_runs = data["all_runs_sorted"]
+    ko_runs = data["all_runs_rejected"]
+    print(f"[stage02b] {len(ok_runs)} OK + {len(ko_runs)} rejected = {len(ok_runs)+len(ko_runs)} runs to plot")
+
+    # Group by AUV → 1 read per AUV bag set covering all runs
+    runs_by_auv = {}
+    for run in ok_runs + ko_runs:
+        # run_id "AUV210_run_00" → AUV210
+        phys = run["run_id"].split("_")[0]
+        runs_by_auv.setdefault(phys, []).append(run)
+
+    pad = args.padding_s
+    for phys_id, runs in runs_by_auv.items():
+        mcap_id = PHYS_TO_MCAP.get(phys_id, phys_id)
+        t_start = min(r["start_epoch"] for r in runs) - pad
+        t_end = max(r["end_epoch"] for r in runs) + pad
+        print(f"  [{phys_id}] reading {mcap_id} bags [{t_start:.0f}..{t_end:.0f}] for {len(runs)} runs")
+        alt, state, hdg = gather_signals_for_auv(bags_root, mcap_id, t_start, t_end)
+        print(f"    {len(alt)} alt pts, {len(state)} state pts, {len(hdg)} hdg pts")
+
+        for run in runs:
+            r_start = run["start_epoch"] - pad
+            r_end = run["end_epoch"] + pad
+            alt_r = [(t, v) for t, v in alt if r_start <= t <= r_end]
+            state_r = [(t, a, m) for t, a, m in state if r_start <= t <= r_end]
+            hdg_r = [(t, v) for t, v in hdg if r_start <= t <= r_end]
+            status = "REJECTED" if run in ko_runs else "OK"
+            reason = run.get("rejected_reason") if status == "REJECTED" else None
+            out_png = out_dir / f"{run['run_id']}.png"
+            plot_run_diag(run, alt_r, state_r, hdg_r, filter_params, status, reason, out_png)
+            print(f"    [{status}] {run['run_id']}.png written")
+
+    # Build index.html + index.json
+    generated_at = datetime.now(timezone.utc).isoformat()
+    index = {
+        "mission": args.mission,
+        "generated_at": generated_at,
+        "filter_params": filter_params,
+        "ok_runs": [r["run_id"] for r in ok_runs],
+        "rejected_runs": [r["run_id"] for r in ko_runs],
+    }
+    (out_dir / "index.json").write_text(json.dumps(index, indent=2))
+    build_html(args.mission, filter_params, generated_at, ok_runs, ko_runs, out_dir)
+    print(f"[stage02b] Done: {out_dir}/index.html")
+
+
+if __name__ == "__main__":
+    main()