The invention discloses an accelerated calculation method of an SSD target detection convolutional neural network. The method comprises the following steps: S1, deleting a target box Box regression layer and a target classification regression layer in an SSD target detection convolutional neural network training stage, and performing Box derivation calculation and target classification calculationat the same time by using a new convolutional layer; S2, adding a Block layer behind the new convolution layer, splitting the Block output by the new convolution layer into two independent Blocks according to the configured channel through the Block layer, and outputting the two independent Blocks to two subsequent related layers in the training stage; S3, deleting a target box Box regression layer and a target classification regression layer in an SSD target detection convolutional neural network derivation stage, and using a new convolution layer to perform Box derivation calculation and target classification calculation at the same time; and S4, completely deleting subsequent related layers in the SSD target detection convolutional neural network derivation stage, and connecting an Object _ Detector layer behind a new convolutional layer.