Identificación y conteo de aceitunas en imágenes digitales tomadas en el olivar mediante morfología matemática y redes neuronales convolucionales

  1. Arturo Aquino 1
  2. Juan Manuel Ponce 1
  3. Borja Millan 1
  4. Diego Tejada-Guzmán 1
  5. José Manuel Andújar 1
  1. 1 Universidad de Huelva
    info

    Universidad de Huelva

    Huelva, España

    ROR https://ror.org/03a1kt624

Book:
XL Jornadas de Automática: libro de actas. Ferrol, 4-6 de septiembre de 2019
  1. Jose Luis Calvo Rolle (coord.)
  2. Jose Luis Casteleiro Roca (coord.)
  3. María Isabel Fernández Ibáñez (coord.)
  4. Óscar Fontenla Romero (coord.)
  5. Esteban Jove Pérez (coord.)
  6. Alberto José Leira Rejas (coord.)
  7. José Antonio López Vázquez (coord.)
  8. Vanesa Loureiro Vázquez (coord.)
  9. María Carmen Meizoso López (coord.)
  10. Francisco Javier Pérez Castelo (coord.)
  11. Andrés José Piñón Pazos (coord.)
  12. Héctor Quintián Pardo (coord.)
  13. Juan Manuel Rivas Rodríguez (coord.)
  14. Benigno Rodríguez Gómez (coord.)
  15. Rafael Alejandro Vega Vega (coord.)

Publisher: Servizo de Publicacións ; Universidade da Coruña

ISBN: 978-84-9749-716-9

Year of publication: 2019

Pages: 818-827

Congress: Jornadas de Automática (40. 2019. Ferrol)

Type: Conference paper

Abstract

Early and accurate yield estimation is a very valued objective for modern agriculture. In the case of oliviculture, it is especially relevant due to the high economic value of its production. This paper presents a methodology aimed at achieving that end. Concretely, it comprises an artificial vision algorithm able to detect those olives that are visible in a digital image of an olive tree, captured directly in the field, at night-time and with artificial illumination. First, the image is preprocessed by means of mathematical morphology techniques and statistical filtering to, from this output, generate a subset of images with high probability of containing an olive. Thus, this preprocessing reduces the search space in a magnitude of 103. Next, these subimages are classified by a convolutional neural network as ‘olive’ or ‘discarded’. From a total of 304,483 subimages, extracted from 21 images, the net correctly classified 98.23% of cases, and gave a coefficient of determination R2 of 0.9875 when facing the number of detected olives to the real one. This achieved accuracy indicates that the found algorithm constitutes a solid step towards the implementation of a future system for early yield estimation of olive orchards