The invention discloses a
branch point detection method for a
tree structure in a
digital image, and the method is a depth
branch point detection model based on a two-stage cascaded convolutional network, i.e., a candidate region segmentation network and a
false detection elimination network. The method comprises the following steps: firstly, extracting a sample with a fixed size from an originalimage to
train a three-dimensional U-shaped segmentation network of an anisotropic
convolution kernel, then inputting an image containing a
tree structure into the trained segmentation network for segmentation to obtain
branch point candidate regions, and taking each point of the candidate regions as a candidate point of a branch point; extracting three 3D image blocks of the candidate points according to three proportions, and calculating the
maximum intensity projection of three views of each 3D image block to form nine corresponding 2D views; meanwhile, inputting the 2D views into the stacks of the five
convolution layers respectively, finally, fusing the features, corresponding to the candidate points, of the 2D views after
convolution, a final branch point detection result is obtained, and the method has the advantages of being low in calculation cost, low in
false detection rate and high in detection efficiency.