An image recognition-based electric speaker includes a U-shaped iron for increasing magnetic field and improving anti-magnetism, a magnet installed on the U-shaped iron for generating magnetic field, a washer installed on the magnet for increasing the magnetic field and magnetic permeability, a voice coil installed on the washer for electrically conducting power, a voice coil paper tube installed on the voice coil, a damper installed around the voice coil for maintaining the magnetic gap and improving the capability of withstanding power, a box holder mounted around the low-frequency suspension edge, a gasket installed at the lower edge of the box holder for sealing the box holder airtightly, a low-frequency suspension edge disposed around the mid-frequency vibrating plate, a high-frequency anti-dust cap installed on the voice coil paper tube, and a mid-frequency vibrating plate disposed around the anti-dust cap. This invention has the effect of reducing criminal activities.