This paper presents work that was done to develop an energy-efficient electoral and frame count system for underwater sea turtle image and video recognition using convolutional neural networks, deep learning framework, and the Python programming language. An underwater sea turtle image recognition program is essential to protect turtles from the threat of bycatch - sea turtles are accidentally caught when fishermen aim for a different type of underwater species. This underwater image recognition system is used to detect the presence of sea turtles, then different kinds of acoustic and light stimuli are used to warn the turtles of approaching danger to reduce bycatch. This image detection system will be placed on a fishing boat to run on a machine at all times (24 hours and 7 days a week). A live video capture from a low-power underwater camera that is attached to the boat will be sent to the image detection system on the machine to analyze the presence of sea turtles in each frame of the video. To lower the computational time and energy of the machine, an energy-efficient electoral and frame count system is implemented on this image detection system.