在我的项目中，我试图估算稻草的切碎细度。在尝试任何CNN技术之前，我也非常想知道OpenCV方法能给我带来多大的好处。 这是这样的图片： This picture shows similar footage which I found on the internet, because the images I'm using are property of the company
到目前为止，我一直在使用不同的图像操作员和不同的设置。多亏了stackoverflow上的一些帖子，我一直在尝试在轮廓检测之前对图像进行预处理，包括模糊，对比度和伽玛调整，重磨，阈值设置，精巧运算符，膨胀和侵蚀...到目前为止，我一直在玩围绕操作员进行切换，更改其顺序和参数并获得了导致以下结果前后的工作流程：
before processing after processing
在预处理工作流中进行编码提取：
image = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
ContourDetection_Variant.show_image(image, "Grayscaled", image_label)
image = cv2.medianBlur(image, 5)
ContourDetection_Variant.show_image(image, "Median Filtered Image", image_label)
# Adjust the brightness and contrast of the picture, alpha is contrast & beta is brightness.
alpha = 3 # Simple contrast control, 1.0 is status quo (alpha value [1.0-3.0])
beta = -100 # Simple brightness control, 0 i status quo (beta value [0-100])
image = cv2.convertScaleAbs(image, alpha=alpha, beta=beta)
ContourDetection_Variant.show_image(image, "Contrast & Brightness adjustment", image_label)
# Sharping method 2: Applying scharpening kernel filter
# https://medium.com/@almutawakel.ali/opencv-filters-arithmetic-operations-2f4ff236d6aa
# Create our shapening kernel, it must equal to one eventually
# kernel_sharpening = np.array([[-1.4142, -1, -1.4142],
# [-1, 10.6569, -1],
# [-1.4142, -1, -1.4142]])
kernel_sharpening = np.array([[-2.8284, -2.2361, -2, -2.2361, -2.8284],
[-2.2361, -1.4142, -1, -1.4142, -2.2361],
[-2, -1, 47.8591, -1, -2],
[-2.2361, -1.4142, -1, -1.4142, -2.2361],
[-2.8284, -2.2361, -2, -2.2361, -2.8284]])
# applying the sharpening kernel to the input image & displaying it.
image = cv2.filter2D(image, -1, kernel_sharpening)
ContourDetection_Variant.show_image(image, "Sharpening method by designing own kernel", image_label)
# Added bordertype REFLECT_101, which extends the image at the borders with pixels within the image in inverse
# order.
image = cv2.GaussianBlur(image, (3, 3), cv2.BORDER_REFLECT_101)
ContourDetection_Variant.show_image(image, "Gaussian Blur", image_label)
retval, image = cv2.threshold(image, 45, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU)
ContourDetection_Variant.show_image(image, "Thresholded", image_label)
image = cv2.erode(image, None, iterations=1)
ContourDetection_Variant.show_image(image, "Erode applied", image_label)
image = cv2.medianBlur(image, 3)
ContourDetection_Variant.show_image(image, "Median Filtered Image", image_label)
retval, image = cv2.threshold(image, 45, 255, cv2.THRESH_BINARY + cv2.THRESH_OTSU)
ContourDetection_Variant.show_image(image, "Thresholded", image_label)
（接下来将获得的图像馈送到OpenCV的findContours中）
挑战是一些吸管很大，我想注册它，但是它们被顶部的一些吸管相交，这使得算法认为这些是单独的吸管。
我的问题很笼统，OpenCV是否可以提供更多有前途的技术（除了寻找轮廓）来实现我的目标？