Fingerlings mass estimation: A comparison between deep and shallow learning algorithms

Authors

Adair da Silva Oliveira Junior, Federal University of Mato Grosso do Sul
Diego André Sant’Ana, Universidade Católica Dom Bosco, Federal Institute of Mato Grosso do Sul
Marcio Carneiro Brito Pache, Universidade Católica Dom Bosco, Federal Institute of Mato Grosso do Sul
Vanir Garcia, Universidade Católica Dom Bosco, Federal Institute of Mato Grosso do Sul
Vanessa Aparecida de Moares Weber, Universidade Católica Dom Bosco, State University of Mato Grosso do Su
Gilberto Astolfi, Federal University of Mato Grosso do Sul
Fabricio de Lima Weber, Federal University of Mato Grosso do Sul, State University of Mato Grosso do Sul
Geazy Vilharva Menezes, Federal University of Mato Grosso do Sul
Gabriel Kirsten Menezes, Federal University of Mato Grosso do Sul
Pedro Lucas França Albuquerque, University of Nebraska Lincoln
Celso Soares Costa, Federal Institute of Mato Grosso do Sul
Eduardo Quirino Arguelho de Queiroz, Universidade Católica Dom Bosco
João Victor Araújo Rozales, Universidade Católica Dom Bosco
Milena Wolff Ferreira, Universidade Católica Dom Bosco
Marco Hiroshi Naka, Federal Institute of Mato Grosso do Sul
Hemerson Pistori, Universidade Católica Dom Bosco

Document Type

Article

Date of this Version

11-1-2021

Citation

A. da Silva Oliveira Junior, D.A. Sant’Ana, M. Carneiro Brito Pache et al. Smart Agricultural Technology 1 (2021) 100020. https://doi.org/10.1016/j.atech.2021.100020

Comments

Open access

Abstract

The paper presents some results regarding the automatic mass estimation of Pintado Real fingerlings, using machine learning techniques to support the fish production process. For this purpose, an image dataset called FISHCV1206FSEG, was created which is composed of 1206 images of fingerlings with their respective annotated masses. Through the fish contours, the area and perimeter were extracted, and submitted to the J48, SVM, and KNN classification algorithms and a linear regression algorithm. The images were also submitted to ResNet50, In- ceptionV3, Exception, VGG16, and VGG19 convolutional neural networks. As a result, the classification algorithm J48 reached an accuracy of 58.2% and a linear regression model capable of predicting the mass of a Pintado Real fingerling with a mean squared error of 1.5 g. The convolutional neural network ResNet50 obtained an accuracy of 67.08%. We can highlight the contributions of this work through the presentation of a methodology to classify the mass of fingerlings in a non-invasive way and by the analyses and comparing results of different machine learning algorithms for classification and regression.