TornadoesTornado frequentlyHook echo occurDeep learning in Japan, similar to the USA. A significant number of tornadoesTornado in Japan originate over water, subsequently making landfall and causing severe damage to structures, vegetation, and human life. To mitigate such disasters, an effective early warning system utilizing weather radar is crucial. The parent storms of tornadoesTornado are often characterized by a vortex signature, specifically a Doppler velocity couplet, which manifests as paired maximum and minimum peaks in Doppler radarDoppler radar observations. However, Doppler velocity data can be significantly contaminated by environmental wind fields, necessitating complex preprocessing techniques. The present study aims to evaluate the efficacy of automatic detection of tornadoTornado-bearing storms based on the hook echoHook echo pattern observed in reflectivity data. A Single Shot Multibox Detector (SSD) was employed as the deep learningDeep learning algorithm. The results demonstrate accurate detection of hook echoHook echo patterns within 0.6 s for most cases, except for low-resolution, full-range surveillance data. The recall and precision metrics are approximately 0.8 and 0.9, respectively.