Upload 2 files

app.py

@@ -0,0 +1,243 @@
+import os, cv2, re, base64
+import numpy as np
+import pandas as pd
+import gradio as gr
+from roboflow import Roboflow
+from openai import OpenAI
+from openpyxl import load_workbook
+
+# ================= CONFIG =================
+ROBOFLOW_API_KEY = "uP19IAi98TqwLvHmNB8V"
+ROBOFLOW_PROJECT = "braker3"
+ROBOFLOW_VERSION = 6
+CONF_THRESHOLD = 0.35
+IOU_THRESHOLD = 0.4
+
+client = OpenAI(api_key=os.getenv("OPENAI_API_KEY"))
+rf = Roboflow(api_key=ROBOFLOW_API_KEY)
+model = rf.workspace().project(ROBOFLOW_PROJECT).version(ROBOFLOW_VERSION).model
+
+# ================= CONSTANTS =================
+SPEC_JP = {
+    "Manufacture Name": "メーカー",
+    "Circuit Name": "回路番号",
+    "Load Name": "負荷名称",
+    "Breaking Capacity": "遮断容量",
+    "AF": "フレーム(AF)",
+    "AT": "トリップ(AT)"
+}
+
+MANUFACTURER_MAP = {
+    "MITSUBISHI ELECTRIC": "三菱電機",
+    "SIEMENS": "SIEMENS",
+    "SCHNEIDER ELECTRIC": "SCHNEIDER ELECTRIC",
+    "ABB": "ABB",
+    "LS ELECTRIC": "LS ELECTRIC"
+}
+
+VALID_BREAKING_CAPACITY = {"6","10","15","25","36","50","65","85"}
+DEFAULT_BREAKING_CAPACITY = "85"   # ← your dataset truth
+
+# ================= IMAGE =================
+def prepare_for_roboflow(img, max_side=1024):
+    h,w = img.shape[:2]
+    scale = min(max_side/max(h,w),1.0)
+    return cv2.resize(img,(int(w*scale),int(h*scale))) if scale<1 else img
+
+def crop(img,x1,y1,x2,y2,pad=20):
+    h,w = img.shape[:2]
+    return img[max(0,y1-pad):min(h,y2+pad), max(0,x1-pad):min(w,x2+pad)]
+
+def enhance(img):
+    g = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
+    g = cv2.resize(g,None,fx=3,fy=3,interpolation=cv2.INTER_CUBIC)
+    clahe = cv2.createCLAHE(2.0,(8,8))
+    return cv2.cvtColor(clahe.apply(g), cv2.COLOR_GRAY2BGR)
+
+def enhance_breaking_capacity(img):
+    g = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
+    g = cv2.resize(g,None,fx=4,fy=4,interpolation=cv2.INTER_CUBIC)
+    g = cv2.adaptiveThreshold(
+        g,255,cv2.ADAPTIVE_THRESH_GAUSSIAN_C,
+        cv2.THRESH_BINARY,31,2
+    )
+    return cv2.cvtColor(g, cv2.COLOR_GRAY2BGR)
+
+def img_to_base64(img):
+    _,buf = cv2.imencode(".jpg",img)
+    return base64.b64encode(buf).decode()
+
+# ================= TEXT =================
+def remove_spaces_only(text):
+    return re.sub(r"\s+", "", str(text)) if text else ""
+
+def extract_digits(text):
+    nums = re.findall(r"\d+",str(text))
+    return nums[-1] if nums else ""
+
+def clean_manufacturer(text):
+    t=text.upper()
+    for k in MANUFACTURER_MAP:
+        if k in t:
+            return k
+    return ""
+
+def normalize_breaking_capacity(text):
+    nums = re.findall(r"\d+",str(text))
+    for n in nums:
+        if n in VALID_BREAKING_CAPACITY:
+            return n
+    return DEFAULT_BREAKING_CAPACITY
+
+# ================= GPT OCR =================
+def gpt_ocr(label,img):
+    if label == "Breaking Capacity":
+        img = enhance_breaking_capacity(img)
+    else:
+        img = enhance(img)
+
+    b64 = img_to_base64(img)
+
+    rules={
+        "Manufacture Name":"Return ONLY manufacturer name in English.",
+        "Circuit Name":"Return EXACT text.",
+        "Load Name":"Return EXACT text.",
+        "AF":"Return ONLY number.",
+        "AT":"Return ONLY number.",
+        "Breaking Capacity":"Return ONLY kA number."
+    }
+
+    r = client.chat.completions.create(
+        model="gpt-5.2",
+        messages=[
+            {"role":"system","content":"Strict OCR engine"},
+            {"role":"user","content":[
+                {"type":"text","text":rules[label]},
+                {"type":"image_url","image_url":{"url":f"data:image/jpeg;base64,{b64}"}}
+            ]}
+        ],
+        temperature=0
+    )
+
+    raw = r.choices[0].message.content.strip()
+
+    if label == "Manufacture Name":
+        return clean_manufacturer(raw)
+    if label in ["Circuit Name","Load Name"]:
+        return remove_spaces_only(raw)
+    if label in ["AF","AT"]:
+        return extract_digits(raw)
+    if label == "Breaking Capacity":
+        return normalize_breaking_capacity(raw)
+
+    return raw
+
+# ================= EXCEL =================
+def normalize_header(s): 
+    return str(s).replace("\n","").replace(" ","")
+
+def find_column(df,keys):
+    for c in df.columns:
+        for k in keys:
+            if k in normalize_header(c):
+                return c
+    return None
+
+def verify_excel(excel,det):
+    wb=load_workbook(excel,data_only=True)
+    ws=wb["MCB"]
+
+    raw=pd.DataFrame([list(r) for r in ws.iter_rows(values_only=True)])
+    raw.dropna(how="all",inplace=True)
+
+    hdr=None
+    for i in range(len(raw)):
+        if "回路" in "".join(map(str,raw.iloc[i].values)):
+            hdr=i; break
+
+    if hdr is None:
+        return pd.DataFrame([["回路番号","", "NO","ヘッダー不明"]],
+            columns=["仕様","検出値","Excelに存在？","備考"])
+
+    df=raw.iloc[hdr+1:].copy()
+    df.columns=raw.iloc[hdr]
+    df.dropna(how="all",inplace=True)
+
+    ccol=find_column(df,["回路番号","回路"])
+    target=None
+    for _,r in df.iterrows():
+        if remove_spaces_only(r[ccol])==det.get("Circuit Name",""):
+            target=r; break
+
+    if target is None:
+        return pd.DataFrame([["回路番号",det.get("Circuit Name",""),"NO","Excelに存在しない"]],
+            columns=["仕様","検出値","Excelに存在？","備考"])
+
+    rows=[]
+    for k,jp in SPEC_JP.items():
+        detv=det.get(k,"")
+        col=find_column(df,[jp.replace("(","").replace(")",""),jp[:2]])
+        excelv=str(target[col]) if col else ""
+
+        if k in ["Circuit Name","Load Name"]:
+            ok=remove_spaces_only(detv)==remove_spaces_only(excelv)
+        elif k=="Manufacture Name":
+            ok=MANUFACTURER_MAP.get(detv,detv)==excelv
+        else:
+            ok=detv==excelv
+
+        rows.append([jp,detv,"YES" if ok else "NO","" if ok else f"Excel値: {excelv}"])
+
+    return pd.DataFrame(rows,columns=["仕様","検出値","Excelに存在？","備考"])
+
+# ================= PIPELINE =================
+def run_pipeline(image,excel):
+    if image is None:
+        return None,pd.DataFrame(),pd.DataFrame(),None
+
+    img=prepare_for_roboflow(image)
+    preds=model.predict(
+        img,
+        confidence=int(CONF_THRESHOLD*100),
+        overlap=int(IOU_THRESHOLD*100)
+    ).json()["predictions"]
+
+    best={}
+    vis=img.copy()
+
+    for p in preds:
+        lab=p["class"]
+        x,y,w,h=map(int,[p["x"],p["y"],p["width"],p["height"]])
+        x1,y1,x2,y2=x-w//2,y-h//2,x+w//2,y+h//2
+        cv2.rectangle(vis,(x1,y1),(x2,y2),(0,255,0),2)
+        c=crop(img,x1,y1,x2,y2)
+        if lab not in best or p["confidence"]>best[lab][0]:
+            best[lab]=(p["confidence"],c)
+
+    det={}
+    rows=[]
+    for lab,(_,c) in best.items():
+        v=gpt_ocr(lab,c)
+        if v:
+            det[lab]=v
+            rows.append([lab,v])
+
+    return vis, pd.DataFrame(rows,columns=["Field","Extracted Text"]), verify_excel(excel,det), "verification_result.xlsx"
+
+# ================= UI =================
+with gr.Blocks(theme=gr.themes.Soft()) as demo:
+    gr.Markdown("# ⚡ Breaker Panel OCR & Verification")
+
+    with gr.Row():
+        img_in=gr.Image(type="numpy",label="Upload Image")
+        img_out=gr.Image(label="Detected Image")
+
+    excel_in=gr.File(label="Upload Excel (MCB)",file_types=[".xlsx",".xlsm"])
+    btn=gr.Button("Run Verification",variant="primary")
+
+    t1=gr.Dataframe(label="OCR Output")
+    t2=gr.Dataframe(label="Verification Result")
+    f=gr.File(label="Download Result")
+
+    btn.click(run_pipeline,[img_in,excel_in],[img_out,t1,t2,f])
+    demo.launch()

requirements.txt

@@ -0,0 +1,9 @@
+gradio
+openai
+roboflow
+opencv-python
+numpy
+matplotlib
+pandas
+openpyxl
+Pillow