Grape picking method based on deep learning and deep information fusion

The invention discloses a grape picking method based on deep learning and deep information fusion. The grape picking method comprises the steps of collecting pictures of grapes; in order to reduce interference of a background on image processing and make an image processing result more accurate in a grape picking point acquisition process, depth information is introduced in a contour detection process, grape target points are marked by comparing the depth information, then values of the marked points are converted into binary images, and a canny algorithm is combined to obtain grape picking points. Interference of environment similar colors and background information is reduced, and the accuracy of contour detection is improved; because grapes and a little background information exist in the ROI of the grapes, a method for calculating the depth in the grape contour is adopted to replace a method for speculating the depth interval where the fruit stems are located by using the average value of all depth values