No abstract

Utilização de redes neurais convolucionais para a detecção de objetos em imagens aéreas adquiridas por drones Use of convolutional neural networks to detect objects in aerial images acquired by drones

No abstract

This work explores the modeling and optimization of the conditions to obtain a set of blue pigments for dyeing reactive cotton, by means of an approach that combines the techniques response surface methodology (RSM) and artificial neural network (ANN). By means of RSM technique the interactions and the effects of the main process variables (factors) on the behavior of coloristic intensity (K.S-1) were investigated. For this, a 26 central composite rotational design (CCRD) was carried out considering the factors temperature, NaCl, Na2CO3, NaOH, processing time and RB5 concentration. The results obtained show that all investigated factors have considerable effect on the behavior of K.S-1. The data produced in the dyeing experiments were used to build and train a Multilayer Perceptron ANN (MLP-ANN) to predict K.S-1, being the input layer of the MLP-ANN designed according to the results achieved by the RSM. The non-linear behavior of dyeing with RB5 was successfully modeled by a three-layer MLP-ANN comprising 6 input neurons, 15 hidden neuros, and 1 output neuron to indicate the value of K.S-1. The results achieved in the performed simulations confirmed the ANN effectiveness to predict K.S-1 values in RB5 the dyeing process, with high coefficient of determination (R2=0.942). The developed approach allowed the composition of a table containing optimized conditions to obtain a set of colors of the blue palette using RB5 dye, varying from sky blue to oxford blue, which will facilitate the assembly of the dyes. Finally, the experiments conducted in this work allowed the development of a computational tool to support the dyeing process, saving chemical inputs and time in cotton dyeing with specific dyestuff.

No abstract