The invention discloses a scattered multi-article robot sorting method in a material frame based on deep study. The method comprises the following steps of selecting a plurality of sortable candidateregions in a range of the material frame; inputting an image and a depth picture of each sortable candidate region into a CNN(convolutional neural network) so as to judge whether the region is the sortable region or not; recording the center of the sortable region into a sortable point; selecting an optional sortable point through a cluster algorithm; calculating a normal vector of the optional sortable point; removing unqualified sortable points by combining with the normal vector, the reachable position of a robot and the anti-corrosion function of the material frame; randomly selecting onesortable point from the rest sortable points; controlling a six-freedom mechanical arm and a suction disc claw to sort articles on the sortable points through the coordinate and the normal vector of the selected sortable points. The excellent feature extraction capability of deep study is used for automatically studying the characteristic of whether scattered articles are sortable or not; the effective dealing with the frequently changed part types, mutual shielding and sorting object uncertainty can be realized.