Update app.py

app.py

@@ -363,6 +363,10 @@ def extract_speed_from_text(full_text: str):
     if not full_text:
         return np.nan
 
+    text = full_text.replace("\xa0", " ")
+    text = re.sub(r"\s+", " ", text)
+
+    # Cas propres
     patterns = [
         r"VMA\s*(\d+(?:[.,]\d+)?)\s*kmh",
         r"VMA\s*(\d+(?:[.,]\d+)?)\s*km/h",
@@ -371,10 +375,20 @@ def extract_speed_from_text(full_text: str):
     ]
 
     for pattern in patterns:
-        m = re.search(pattern, full_text, flags=re.I)
+        m = re.search(pattern, text, flags=re.I)
         if m:
             return float(m.group(1).replace(",", "."))
 
+    # Cas OCR cassé observé : "A Mmh14Vk"
+    m = re.search(r"Mmh\s*(\d+(?:[.,]\d+)?)\s*Vk", text, flags=re.I)
+    if m:
+        return float(m.group(1).replace(",", "."))
+
+    # Variante encore plus souple
+    m = re.search(r"[Mm]\w{0,3}\s*(\d+(?:[.,]\d+)?)\s*[Vv][Kk]", text)
+    if m:
+        return float(m.group(1).replace(",", "."))
+
     return np.nan
 
 def parse_zebris_pdf(uploaded_pdf):