图块拼接BetaTest

README.md

@@ -125,3 +125,21 @@
 |   识别结果   | ![image-20220721203004305](https://raw.githubusercontent.com/Karllzy/imagebed/main/img/image-20220721203004305.png) |
 | 模型约束情况 | ![image-20220722095620511](https://raw.githubusercontent.com/Karllzy/imagebed/main/img/image-20220722095620511.png) |
 
+## 图像的对齐
+
+引入了RGB和光谱图像的原因，这里牵扯到图像对齐的问题。
+
+这里可以看到对齐的结果：
+
+![Figure_1](https://raw.githubusercontent.com/Karllzy/imagebed/main/img/Figure_1.png)
+
+根据这张图片的换算结果可以得知光谱图像比RGB图像超前了69个像素，大概2.02厘米的样子。
+
+![Figure_2](https://raw.githubusercontent.com/Karllzy/imagebed/main/img/Figure_2.png)
+
+根据这张图片，可以得知，图像上下偏差是2.3厘米
+
+
+![Figure_4](https://raw.githubusercontent.com/Karllzy/imagebed/main/img/Figure_4.png)
+
+这张图片里的上下偏差则达到了2.6厘米左右。

config.py

@@ -30,5 +30,9 @@ class Config:
     threshold_s = 175
     rgb_size_threshold = 4
 
-    # Mask parameter
+    # mask parameter
-    target_size = (1024, 1024)  # (Width, Height) of mask
+    target_size = (256, 1024)  # (Width, Height) of mask
+
+    # save part
+    offset_vertical = -70
+

main.py

@@ -25,12 +25,17 @@ def main():
         os.mkfifo(rgb_fifo_path, 0o777)
     if not os.access(mask_fifo_path, os.F_OK):
         os.mkfifo(mask_fifo_path, 0o777)
-    if not os.access(rgb_mask_fifo_path, os.F_OK):
+
-        os.mkfifo(rgb_mask_fifo_path, 0o777)
+    # 进行补偿buffer的开启
+    if Config.offset_vertical < 0:
+        # 纵向的补偿小于0，那就意味着光谱图要上移才能补上，那么我们应该补偿SPEC相机的全 0 图像
+        conserve_part = np.zeros((abs(Config.offset_vertical) // 4, Config.nRows, Config.nBands))
+    elif Config.offset_vertical > 0:
+        # 纵向的补偿小于0，说明光谱图下移才能补上去，那么我们就需要补偿RGB相机的全 0 图像
+        conserve_part = np.zeros(abs(Config.offset_vertical), Config.nRgbRows, Config.nRgbBands)
     while True:
         fd_img = os.open(img_fifo_path, os.O_RDONLY)
         fd_rgb = os.open(rgb_fifo_path, os.O_RDONLY)
-
         # spec data read
         data = os.read(fd_img, total_len)
         if len(data) < 3:
@@ -40,7 +45,6 @@ def main():
         else:
             data_total = data
         os.close(fd_img)
-
         # rgb data read
         rgb_data = os.read(fd_rgb, total_rgb)
         if len(rgb_data) < 3:
@@ -51,23 +55,34 @@ def main():
         else:
             rgb_data_total = rgb_data
         os.close(fd_rgb)
-
         # 识别
         t1 = time.time()
-        img_data = np.frombuffer(data_total, dtype=np.float32).reshape((Config.nRows, Config.nBands, -1)) \
+        img_data = np.frombuffer(data_total, dtype=np.float32).reshape((Config.nRows, Config.nBands, -1))\
-            .transpose(0, 2, 1)
+                     .transpose(0, 2, 1)
         rgb_data = np.frombuffer(rgb_data_total, dtype=np.uint8).reshape((Config.nRgbRows, Config.nRgbCols, -1))
+        if Config.offset_vertical < 0:
+            # 纵向的补偿小于0，那就意味着光谱图要上移才能补上，那么我们应该补偿SPEC相机的全 0 图像
+            new_conserve_part, real_part = img_data[:abs(Config.offset_vertical) // 4, ...],\
+                                       img_data[abs(Config.offset_vertical) // 4:, ...]
+            img_data = np.concatenate([real_part, conserve_part], axis=0)
+            conserve_part = new_conserve_part
+        elif Config.offset_vertical > 0:
+            # 纵向的补偿小于0，说明光谱图下移才能补上去，那么我们就需要补偿RGB相机的全 0 图像
+            new_conserve_part, real_part = rgb_data[:abs(Config.offset_vertical), ...],\
+                                           rgb_data[abs(Config.offset_vertical):, ...]
+            rgb_data = np.concatenate([real_part, conserve_part], axis=0)
+            conserve_part = new_conserve_part
         # 光谱识别
         mask_spec = spec_detector.predict(img_data)
         # rgb识别
         mask_rgb = rgb_detector.predict(rgb_data)
         # 结果合并
-        # mask_result = (mask_spec | mask_rgb).astype(np.uint8)
+        mask_result = (mask_spec | mask_rgb).astype(np.uint8)
 
         # control the size of the output masks
         masks = [cv2.resize(mask.astype(np.uint8), Config.target_size) for mask in [mask_spec, mask_rgb]]
         # 写出
-        output_fifos = [mask_fifo_path, rgb_mask_fifo_path]
+        output_fifos = [mask_fifo_path, ]
         for fifo, mask in zip(output_fifos, masks):
             fd_mask = os.open(fifo, os.O_WRONLY)
             os.write(fd_mask, mask.tobytes())
@@ -82,7 +97,6 @@ if __name__ == '__main__':
     rgb_fifo_path = "/tmp/dkrgb.fifo"
 
     mask_fifo_path = "/tmp/dkmask.fifo"
-    rgb_mask_fifo_path = "/tmp/dkmask_rgb.fifo"
     # 主函数
     main()
     # read_c_captures('/home/lzy/2022.7.15/tobacco_v1_0/', no_mask=True, nrows=256, ncols=1024,

utils.py

@@ -140,7 +140,6 @@ def size_threshold(img, blk_size, threshold):
     return mask
 
 
-
 if __name__ == '__main__':
     color_dict = {(0, 0, 255): "yangeng", (255, 0, 0): "bejing", (0, 255, 0): "hongdianxian",
                   (255, 0, 255): "chengsebangbangtang", (0, 255, 255): "lvdianxian"}