Design approach of a trout egg sorting machine based on convolutional neural networks with water conditions monitoring

Today's rainbow trout egg farming process is still a manual labor. This costs a great amount of resources and time for the fish farmers that take care of this specie. Along the Peruvian rainbow trout egg value chain, the most critical aspect is the death rate that results in lower productivity and, in consequence, lower income for this type of activity. The following article describes the design of two systems: a trout egg sorting machine and a trout egg incubator. The first one includes microcontrollers, electric pumps, computer vision features, sensors, valves and pipes. The other one consists of pipes, valves and sensors. The main aim of this current project is to help rainbow trout egg farmers by providing them a modern way to separate dead fish eggs from the living ones through a Convolutional Neural Network, which is a multilayered architecture and allows to focus on the most relevant features in the image processing. Then, keep the amount of live fish eggs in an appropriate incubator, all together with a group of sensors located around the eggs in order to preserve them for the next step in the rainbow trout hatching. The results showed that the data from the water parameters could be visualize through an user interface and the trout egg classification made by the trained model achieved a maximum accuracy of 95% during the training process.