Abstract: Mining electric shovel is an important equipment of open pit mine production, parts on bucket fall off frequently during the work of shovel, and it is not easy to be found in time. If these parts enter the downstream crushing link along with the ore, it is easy to cause damage to crusher, resulting in shutdown of the production line for maintenance and huge economic losses. At present, the monitoring of bucket parts falling off mainly depends on the visual judgment of the shovel driver, it is inefficient and inaccurate, which distracts the driver. Based on deep learning and machine vision technology, a method for monitoring is studied to complete the detection of the bucket parts falling off, and a corresponding detection system is designed. Based on SSD preprocessing model, the model is optimized and trained. An industrial camera is installed on the bucket boom of the electric shovel, images are collected and input into the touch control machine in the cab to call the training model, and machine vision technology is used to detect whether the bucket parts fell off. In Qidashan Iron Mine of Ansteel Mining, the accuracy of the detection method is verified for half a year, and the accuracy rate reached 90%. The results show that the method can detect the parts falling off in real time, solve the problem that the traditional method can only be identified by human eyes, reduce labor intensity, reduce unnecessary downtime, and ensure continuous production.