我需要在sobel运算符之后发布图像，但是ROS消息不支持浮点数。那么如何发布边缘检测消息？

pub = rospy.Publisher('detector',Image,queue_size=10)

def image_callback(msg):
    print("Received an image")

    try:
        cv2_img = bridge.imgmsg_to_cv2(msg,"bgr8")
        #print(msg)
        sobel(cv2_img)
    except CvBridgeError,e:
        print(e)

def sobel (photo):


    gray = cv2.cvtColor(photo,cv2.COLOR_BGR2GRAY)
    im_original = np.array(gray)

    #im_original = np.array(photo.convert('L'))
    imy = np.zeros(im_original.shape)
    imx = np.zeros(im_original.shape)

    # Run the Sobel operator

    filters.sobel(im_original,1,imx,cval=0.0)  # axis 1 is x
    filters.sobel(im_original,imy,cval=0.0) # axis 0 is y

    magnitude = np.sqrt(imx**2+imy**2)
    #cv2.imshow('lol',magnitude)
    print(magnitude.shape)
    #imgMsg = bridge.cv2_to_imgmsg(np.array[magnitude],"bgr8")
    message = bridge.cv2_to_imgmsg(magnitude,"64FC1")  

    pub.publish(message)

    # Construct the plot


def main():
    rospy.init_node('image_listener')

    image_topic = "/usb_cam/image_raw"

    rospy.Subscriber(image_topic,image_callback)

    rospy.spin()


if __name__ == '__main__':
    main()

如果我输入“ bgr8”，则表示指定为bgr8的编码，但是图像具有不兼容的类型64FC1。 我该如何克服这个问题？

Received an image
[[  0.           4.           1.41421356 ...,0.           3.16227766
    4.24264069]
 [  0.           4.           3.16227766 ...,0.           1.41421356
    3.16227766]
 [  0.           3.16227766   4.47213595 ...,0.           0.           0.        ]
 ...,[  3.16227766   4.47213595   2.         ...,51.86520992  30.46309242
   11.66190379]
 [  3.16227766   4.47213595   2.         ...,57.00877125  39.2173431
   15.23154621]
 [  1.41421356   4.47213595   2.         ...,59.22837158  45.54119015
   18.11077028]]