Tiny target detection method oriented to behavior monitoring of driver of special vehicle

The invention relates to a tiny target detection method for behavior monitoring of a driver of a special vehicle, and belongs to the field of computer vision. The method specifically comprises the steps that firstly, simulation data highly similar to real data are generated through iterative diffusion operation by using a Diffusion network through a Markov chain mode of front and back derivation, and a tiny target in a preliminarily processed image is pasted to any position through a specific instance copying and pasting technology to generate a composite image containing rich tiny target information; the dual-path parallel distillation learning network performs distillation to obtain a series of Student networks in different application scenes to form network spaces suitable for different special vehicles; and finally, realizing matching of the special vehicle and the Student network in different scenes by using a multi-source multi-target optimization knowledge matching network. The method can be matched wi