Hot subdwarf stars, located at the end of the blue horizontal branch, have a burning helium core, and are crucial for studying stellar structure. Currently, the number of known hot subdwarf star is relatively small. Many studies apply machine learning classification to search for hot subdwarf stars based on high quality spectra. With the accumulated image data in the digital sky surveys, we proposed the Hot Subdwarf Detector (HsdDet), a novel multiscale object detection algorithm designed to directly locate hot subdwarf stars' coordinates in Sloan Digital Sky Survey (SDSS) images. We applied the HsdDet algorithm to some photometric images of SDSS, and detected 29,695 candidates, whose reduced proper motions are primarily concentrated between 5 and 20mas/yr. Most of the candidates have distances of 1.5~7.5kpc, while some candidates can be as far as 20kpc or even more. Candidates' colors are -2.5<u-g<2.0, -3.5<g-r<2.0, -2.5<r-i<2.5, and -3.0<i-z<3.0. We crossmatched the candidates catalog with Large Sky Area Multi-Object Fiber Spectroscopic Telescope and obtained 3823 spectra. The confidence measures the probability of a candidate being a hot subdwarf star. With the confidence threshold being 0.5, 263 new hot subdwarf stars were confirmed by the spectral fitting method, and the purity is 6.9%. When we set confidence threshold at 0.9, the purity is 13.5%. We discussed the selection bias in the search region and released the hot subdwarf stars catalog. These results have the potential to expand the population of hot subdwarf stars.