The invention discloses a meat
production line automatic segmentation and positioning method based on three-dimensional
visual identification, and the method comprises the steps: synchronously collecting a
color image and a depth image, carrying out the
spatial calibration and alignment, and obtaining a
data set; performing
color image channel difference and depth
image gradient calculation to generate a
tissue level response volume; multi-
modal input is sent to an improved MaskDINO network, prior query and attention are carried out, and a segmentation
mask is output; performing spatial continuity and stability analysis on the segmentation
mask to generate segmentation stability parameters; in combination with response volume and stability parameters, constructing a
cutting potential field, and performing three-dimensional search positioning
cutting and grabbing; and calculating position and attitude parameters, outputting a control instruction, and driving segmentation and
positioning equipment. According to the method, stable segmentation and accurate automatic positioning of a complex meat target in a meat
production line are realized by fusing three-dimensional
visual identification of
tissue level prior and an improved instance segmentation method.