Skip to main content
Disciplines
Automotive Business IT + Informatics Construction + Real Estate Electrical Engineering + Electronics Energy + Sustainability Insurance + Risk Finance + Banking Management + Leadership Marketing + Sales Mechanical Engineering + Materials
Events
DE
EN
Books
Journals
Topic Page
Marketing
Start single access now
Access for companies
EXTENDED SEARCH
Log in
Top

2023 | Book

Design of a Segmentation and Classification System for Seed Detection Based on Pixel Intensity Thresholds and Convolutional Neural Networks Read chapter Read first chapter

Applications of Computational Intelligence

5th IEEE Colombian Conference, ColCACI 2022, Cali, Colombia, July 27–29, 2022, Revised Selected Papers

Editors: Alvaro David Orjuela-Cañón, Jesus Lopez, Julian David Arias-Londoño, Juan Carlos Figueroa-García

Publisher: Springer Nature Switzerland

Book Series : Communications in Computer and Information Science

Part of: Springer Professional "Wirtschaft+Technik" , Springer Professional "Technik" , Springer Professional "Wirtschaft"

Login to get access
insite
SEARCH

About this book

​This book constitutes the refereed proceedings of the 5th IEEE Colombian Conference on Applications of Computational Intelligence, ColCACI 2022, held in Cali, Colombia during July 27–29, 2022.

The 7 extended papers included in this book were carefully reviewed and selected from 38 submissions. They were organized in topical sections as follows: ​Design of a segmentation and classification system for seed detection based on pixel intensity thresholds and convolutional neural networks.

Table of Contents

Frontmatter
Design of a Segmentation and Classification System for Seed Detection Based on Pixel Intensity Thresholds and Convolutional Neural Networks
Abstract
Due to the computational power and memory of modern computers, computer vision techniques and neural networks can be used to develop a visual inspection system of agricultural products to satisfy product quality requirements. This chapter employs artificial vision techniques to classify seeds in RGB images. As a first step, an algorithm based on pixel intensity threshold is developed to detect and classify a set of different seed types, such as rice, beans, and lentils. Then, the information inferred by this algorithm is exploited to develop a neural network model, which successfully achieves learning classification and detection tasks through a semantic-segmentation scheme. The applicability and satisfactory performance of the proposed algorithms are illustrated by testing with real images, achieving an average accuracy of 92% in the selected set of classes. The experimental results verify that both algorithms can directly detect and classify the proposed set of seeds in input RGB images.
Oscar J. Suarez, Edgar Macias-Garcia, Carlos J. Vega, Yersica C. Peñaloza, Nicolás Hernández Díaz, Victor M. Garrido
Classification of Focused Perturbations Using Time-Variant Functional Connectivity with rs-fmri
Abstract
Focal disturbances in the cerebral activity modulated by transcranial magnetic stimulation (TMS) produce alterations in brain connectivity at the global level. Those effects can be studied using time-varying functional connectivity (TVFC) based on functional magnetic resonance imaging at rest (rs-fMRI). The characteristics of these alterations could be modeled using machine learning algorithms for patient classification. This study used hidden Markov models (HMM) to evaluate temporal variations in functional connectivity after stimulation of two different brain areas (Frontal and Occipital). We modeled the dynamics of 15 resting-state networks in 12 states by calculating the fractional occupancy, mean lifetime, and interval time of each state. We then compared the difference between fMRI sessions, PRE, and POST-stimulus, observing significant differences for both conditions, especially after frontal stimulation. Finally, generative models based on HMM were trained, to classify PRE-stimulus and Occipital stimulus with an accuracy of 83%, PRE-stimulus and Frontal stimulus with an accuracy of 85%, and Occipital and Frontal stimulus with 65% accuracy. This finding could be extended to the characterization of pathologies where local disturbances have a global impact on functional connectivity, such as Epilepsy.
Catalina Bustamante, Gabriel Castrillón, Julián Arias-Londoño
Escherichia coli: Analysis of Features for Protein Localization Classification Employing Fusion Data
Abstract
Machine learning models can be used for relevance of features in classification systems. The interest in protein analysis based on biomolecular information has rapidly grown. In this case a comparison of two sources of this information was employed to determine protein localization in Escherichia coli cells. Models as support vector machines, artificial neural networks and random forest were compared for the prediction of protein localization. The sources of data used to train the models were the information from targeting signal and protein sequences, for determining the localization sites of the protein. A third scenario with a fusion of both sources of data was employed. Four classes were established according to the subcellular localization of the protein: cytoplasm, periplasmatic space, outer and inner membranes. Results reached values between 77% and 92% in terms of balanced accuracy. The models with better performance were based on random forest and support vector machines. In terms of features, the first source, where targeting signal was employed, was the one with best performance associated to relevance for the classification.
Alvaro David Orjuela-Cañon, Diana C. Rodriguez, Oscar Perdomo
Artificial Bee Colony-Based Dynamic Sliding Mode Controller for Integrating Processes with Inverse Response and Deadtime
Abstract
A strategy for optimizing the settings of a dynamical sliding mode controller using an artificial bee colony optimization algorithm is proposed in this paper. The performance of the obtained controller is then evaluated and compared to that of a conventional PID and a dynamical sliding mode controller that has been optimized through a heuristics-based strategy by simulating two integrating linear systems with dead time and inverse response. By utilizing the suggested strategy, it was possible to increase both performance indices and transient characteristics.
Jorge Espin, Sebastian Estrada, Diego S. Benítez, Oscar Camacho
Optimizing a Dynamic Sliding Mode Controller with Bio-Inspired Methods: A Comparison
Abstract
In the past few years, bio-inspired optimization algorithms have shown to be an excellent way to solve a wide range of complex computing problems in science and engineering. This paper compares bio-inspired algorithms to better understand and measure how well they find the best tuning parameters for a Dynamic Sliding Mode Control for integrating systems with an inverse response and dead time. The comparison includes four bioinspired algorithms: particle swarm optimization, artificial bee colony, ant colony optimization, and genetic algorithms. It shows how they can improve the performance of the controller by looking for the best tuning parameter solutions. The parameters of each algorithm affect the searching mechanism in different ways, and these effects were tested in two simulated systems. Ant colony optimization is much better than other algorithms at finding the best answers to our problems.
Jorge Espin, Sebastian Estrada, Diego S. Benítez, Oscar Camacho
A Robust Controller Based on LAMDA and Smith Predictor Applied to a System with Dominant Time Delay
Abstract
This document presents a LAMDA (Learning Algorithm for Multivariable Data Analysis) Sliding-Mode Control LSMC applied to a pH neutralization reactor with dominant dead time. Due to the non-linearity generated by the dead time, the application of an additional Smith Predictor structure is proposed to improve the system’s response when reference changes and disturbances occur around the operating point in which the system works. The controller is validated through different simulations in which it is evident that the proposed approach is stable in controlling the pH neutralization reactor.
Luis Morales, Oscar Camacho, Paulo Leica
Comparison of the Performance of Two Neural Network Models with Parameter Optimization for the Prediction of the Bancolombia Share Price
Abstract
The following article explores the characteristics of a NARX neural network, which is used to predict the time series of Bancolombia’s stock. The search for the best characteristics of the network is carried out with a heuristic model, which uses a genetic algorithm to vary the different parameters of the network, the solutions explored are 5000 in each independent experiment, out of a total of 15 independent experiments that reduces the probability of falling into local minima.
Juan Sebastian Castillo Amaya, Juan Pablo Ramírez Villamil, Andrés Eduardo Gaona Barrera
Backmatter
Metadata
Title
Applications of Computational Intelligence
Editors
Alvaro David Orjuela-Cañón
Jesus Lopez
Julian David Arias-Londoño
Juan Carlos Figueroa-García
Copyright Year
2023
Electronic ISBN
978-3-031-29783-0
Print ISBN
978-3-031-29782-3
DOI
https://doi.org/10.1007/978-3-031-29783-0

Premium Partner

    Image Credits