pd9427

vision/charge_3cls/charge_cls_rknn.py

@@ -0,0 +1,131 @@
+import os
+import cv2
+import numpy as np
+
+from collections import deque
+
+class StableClassJudge:
+    """
+    连续三帧稳定判决器：
+    - class0 / class1 连续 3 帧 -> 输出
+    - class2 -> 清空计数，重新统计
+    """
+
+    def __init__(self, stable_frames=3, ignore_class=2):
+        self.stable_frames = stable_frames
+        self.ignore_class = ignore_class
+        self.buffer = deque(maxlen=stable_frames)
+
+    def reset(self):
+        self.buffer.clear()
+
+    def update(self, class_id):
+        if class_id == self.ignore_class:
+            self.reset()
+            return None
+
+        self.buffer.append(class_id)
+
+        if len(self.buffer) < self.stable_frames:
+            return None
+
+        if len(set(self.buffer)) == 1:
+            stable_class = self.buffer[0]
+            self.reset()
+            return stable_class
+
+        return None
+
+# ---------------------------
+# 三分类映射
+# ---------------------------
+CLASS_NAMES = {
+    0: "插好",
+    1: "未插好",
+    2: "有遮挡"
+}
+
+# ---------------------------
+# RKNN 全局实例（只加载一次）
+# ---------------------------
+_global_rknn = None
+
+def init_rknn_model(model_path):
+    from rknnlite.api import RKNNLite
+    global _global_rknn
+    if _global_rknn is not None:
+        return _global_rknn
+
+    rknn = RKNNLite(verbose=False)
+    ret = rknn.load_rknn(model_path)
+    if ret != 0:
+        raise RuntimeError(f"Load RKNN failed: {ret}")
+
+    ret = rknn.init_runtime(core_mask=RKNNLite.NPU_CORE_0)
+    if ret != 0:
+        raise RuntimeError(f"Init runtime failed: {ret}")
+
+    _global_rknn = rknn
+    print(f"[INFO] RKNN 模型加载成功：{model_path}")
+    return rknn
+
+
+# ---------------------------
+# 预处理（输入 uint8，RKNN 内部转 float32）
+# ---------------------------
+def resize_stretch(image, size=640):
+    return cv2.resize(image, (size, size))
+
+def preprocess_image_for_rknn(img, size=640):
+    # 输入必须是 uint8 [0,255]，即使模型是 float32
+    img_resized = resize_stretch(img, size=size)
+    img_rgb = cv2.cvtColor(img_resized, cv2.COLOR_BGR2RGB)
+    input_tensor = np.expand_dims(img_rgb, axis=0).astype(np.uint8)  # NHWC, uint8
+    return np.ascontiguousarray(input_tensor)
+
+
+# ---------------------------
+# 单次 RKNN 推理（三分类，float32 模型）
+# ---------------------------
+def rknn_classify_preprocessed(input_tensor, model_path):
+    rknn = init_rknn_model(model_path)
+    outs = rknn.inference([input_tensor])
+    
+    # 直接得到 logits
+    probs = outs[0].flatten().astype(np.float32)  # shape: (3,)
+    class_id = int(np.argmax(probs))
+    return class_id, probs
+
+
+# ---------------------------
+# 单张图片推理
+# ---------------------------
+def classify_single_image(frame, model_path, size=640):
+    if frame is None:
+        raise FileNotFoundError("输入帧为空")
+
+    input_tensor = preprocess_image_for_rknn(frame, size=size)
+    class_id, probs = rknn_classify_preprocessed(input_tensor, model_path)
+    class_name = CLASS_NAMES.get(class_id, f"未知类别 ({class_id})")
+
+    return {
+        "class_id": class_id,
+        "class": class_name,
+        "score": round(float(probs[class_id]), 4),
+        "raw": probs.tolist()
+    }
+
+
+# ---------------------------
+# 示例调用
+# ---------------------------
+if __name__ == "__main__":
+    model_path = "charge0324.rknn" 
+    image_path = "class2.jpg"
+
+    frame = cv2.imread(image_path)
+    if frame is None:
+        raise FileNotFoundError(f"无法读取图片：{image_path}")
+
+    result = classify_single_image(frame, model_path)
+    print("[RESULT]", result)