Maestría en Electrónica y Automatización

Permanent URI for this collectionhttp://repositorio.uta.edu.ec/handle/123456789/35313

Browse

Filters

20221

Settings

Author: search.filters.author.Cañar Yumbolema, Willam Patricio

Search Results

Now showing 1 - 1 of 1
  • No Thumbnail Available
    Item
    Controlador para alimentación de peces empleando Deep Learning
    (Universidad Técnica de Ambato. Facultad de Ingeniería en Sistemas, Electrónica e Industrial. Maestría en Electrónica y Automatización, 2022) Cañar Yumbolema, Willam Patricio; Galarza Zambrano, Eddie Egberto
    This degree project consists of the design and implementation of a controller for fish feeding using Deep Learning for which a Raspberry minicomputer and a lowcost Web camera were used, this significantly reduced the investment for project development. The main objective consists in the creation of a manually labeled dataset (set of images) of several zebra-type fish (Danio Rerio) located inside a fish tank uniformly illuminated by white LED light. In this case, it was decided to use two videos with the presence of 4 and 6 fish, respectively. Through the use of a computational algorithm, the sequence of images of the fish where their movements can be identified were obtained. This information is used to train a convolutional network using the ACF (Aggregated Channel Characteristics) image object detection algorithm. Once the location of the fish inside the fish tank is determined, the feeling of the school is identified through the implementation of three zones, that is, the developed algorithm will allow knowing if the fish are in a satisfaction zone, a normal zone or a normal zone. feeding. Finally, the FuzzySN, FuzzySH and FuzzyNH indices contain the feeling of the fish and are the inputs of a fuzzy controller which in turn contains the feeding rules based on the natural behavior of the fish; In this way, the developed system is capable of feeding the fish automatically. The minimum identification error reached was 29.5%, but the identification of the behavior of the school of fish had a success rate of 100%. The test carried out to validate the algorithm was given for a case of manual feeding by an operator, where the system was able to correctly identify the feeling of the school as satisfied.