Estimating the position of a surgical instrument and identifying its type are important first steps in developing surgical assistance and automation technology. However, there are few reports on object detection technology using a surgical microscope in the field of ophthalmology. Some of the major challenges in position estimation and tool recognition under microsurgery in this field are that the target area is narrow, the image may be distorted through refraction at the air–liquid interface, and the angle of view is often enlarged or reduced as needed by the surgeon. To address these challenges, we applied a single-shot multibox detector (SSD) technique to determine the position and type of an instrument during cataract surgery. SSD is an object detection technique with superior accuracy and processing speed to existing deep-learning-based detection methods. Using this method, we detected two major surgical tools in images during cataract surgery and obtained a mean average precision of 0.75. Our results show that it is possible to recognize instruments inserted in the eye and estimate their positions.