Current interests, active and past research.

Tree Detection and Classification for Forest Monitoring

Motivation: Detection and classification of trees from remote sensing data are usually performed on multispectral and hyperspectral images. Another sensor is the adoption of Light Detection And Ranging (LiDAR) data.

Gap: Despite the comparatively lower cost and higher spatial resolution, few studies focused on images captured by Red-Green-Blue (RGB) sensors.

Solution: In this context, we have proposed a convolutional neural network to measure the forestry inventory by detecting and classifying tree species. We can also estimate how healthy the tree is. Some results are shown over conifer trees in the first two figures (use the arrows).

Role
Principle investigator
Institutions
Getulio Vargas Foundation (FGV)
Years
2023

A Deep CNN-based regressor for measuring the impact on fish species using otolith images

Motivation: Climate change has been affecting the biodiversity in the oceans.

Gap: The estimation of the age mainly using otoliths is entirely performed by visual analysis of biologists and ecologists.

Solution: A CNN-type architecture for regression was developed by modifying the output layers and then calculating the probability to belong to a certain age.

Role
Principle investigator
Institutions
Dalhousie University, Fisheries and Oceans Canada, Bedford Institute of Oceanography
Years
2020-2021

NanoImage Analyzer

Motivation: Complex networks have been widely used in science and technology because of their ability to represent several systems. One of these systems is found in Biochemistry, in which the synthesis of new nanoparticles is a hot topic.

Gap: The interpretation of experimental results in the search for new nanoparticles poses several challenges. A property of recurrent interest is the agglomeration of particles.

Solution: This research introduced an novel approach that uses complex networks to detect and describe nanoparticle agglomerates (clusters). In this approach, each detected particle in an image corresponds to a vertice and the distances between the particles define a criterion for creating edges. Edges are created if the distance is smaller than a radius of interest. Once this network is built, we calculate several discrete measures able to reveal the most outstanding agglomerates in a nanoparticle image.

Role
Principle investigator
Institutions
University of São Paulo, Federal University of Mato Grosso do Sul, Freie Universität Berlin
Years
2015 - 2017
Website
http://nanoanalyzer.icmc.usp.br

DropLeaf - Spraying Meter

Motivation: Pesticide application has been heavily used in the cultivation of major crops, contributing to the increase in crop production over the past decades.

Gap: The overdosage of pesticides can contaminate food, harm human health, pollute the environment and, can lead to severe economic losses. In contrast, under dosage might cause pest and/or weed resistance or behavioral avoidance.

Solution: We proposed the DropLeaf - Spraying Meter to easily control and assess the spraying of pesticides. Before spraying, experts place several cards in target areas and then automatic assessment is performed using this application.

Role
Principle investigator
Institutions
University of São Paulo, Federal University of Mato Grosso do Sul, Dalhousie University
Years
2015 - 2017
Website
http://dropleaf.icmc.usp.br
# Downloads
> 6,500

BioLeaf - Foliar Analysis

Motivation: Soybean is one of the ten greatest crops in the world, answering for billion-dollar businesses every year. This crop suffers from insect herbivory that costs millions from producers.

Gap: Current methods to measure foliar damage are expensive and dependent on laboratory facilities, in some cases, depending on complex devices that may cost $15,000.

Solution: BioLeaf is a non-destructive imaging smartphone application to estimate foliar loss in leaves with or without border damage. It is applicable not only to soy, but also to different crops such as cotton, bean, potato, coffee, and vegetables.

Role
Principle investigator
Institutions
University of São Paulo, Dom Bosco Catholic University, Federal University of Mato Grosso do Sul
Years
2013 - 2015
Website
http://bioleaf.icmc.usp.br
# Downloads
> 43,000

PEx-Image - Data Analysis and Visualization of Image Collections

Motivation: Projection Explorer for Images (PEx-Image) is a tool used to create and explore visual representations of image collections, helping the user to understand their contents.

Gap: Determining the accuracy achieved by visual feature representation is a challenge since there is no guidance during the process, a complete black box. In particular to my research, image datasets were explored using many ways of description - SIFT-based attributes, color and texture attributes. Several handy-feature methods were used with classifiers that could reflect different properties of the image dataset under investigation.

Solution: In my master thesis, I proposed a novel approach to measure the quality of the visual representation by means of cluster validity instead of using image classification metrics and subjective analysis only. Here, explored visual representation relied particularly on MDS was able to preserve distances in the metric space. In addition, PCA and K-PCA were evaluated as a pre-processing step before data projection to the metric space. On the left side, the first two figures show PCA-like embeddings with MDS visualization.

Role
Participant
Institution
University of São Paulo
Years
2008 - 2010

LARVIC - Computer Vision System for Automatic Larvae Mortality

Motivation: Over the last decades, research on disease vector control has received paramount attention as tropical diseases can cost billions in health care programs. Particularly in Central, South America and some Caribbean countries, only the Aedes aegyptti mosquito is the primary vector for transmission of Dengue, Chikungunya and Zika virus.

Gap: Typically, the insecticides applied in entire communities are costly. Environmentally friendly biodegradable insecticides from local plants have been widely investigated by many laboratories as an intervention measure.

Nonetheless, larval mortality tracking is counted within 24 hours of exposure. Additionally, some larvae can stay stationary for hours, misleading annotators.

Solution: We developed a computational system for larvicidal tests. Video assessment has enabled researchers to test several compounds and different concentrations at once. Instead of measuring a unique sample for each experiment, we could create a grid of samples and measure up to 12 at once.

Role
Participant
Institution
Dom Bosco Catholic University
Years
2007 - 2008

TOPOLINO - Computer Vision System for Automatic Rodent Behaviour Analysis

Motivation: An essential step in drug development relies on measuring the effect on animals (Topolino was strictly used for rodents). In the video on the side-left, you can see three behavioural tests: first open field, followed by plus-maze and water-maze.

Gap: Manual annotation and visual observation could impair long experiments and worsen with multiple mice in the arena.

Solution: In my undergraduate project, it was developed a fully-automated computer vision system for extracting specific features, to name a few, path trajectory after hours of experiment, speed, acceleration, and distance over each part of the arena. It is worth saying that such features are crucial for studies in many areas such as neuroscience, pharmacology, physiology and psychology.

The Topolino system was built-in for running on multi-platforms (Java-based) and could capture and save videos using RGB cameras. This system was the winner of two awards relating to innovation and was also assessed in many laboratories, 17 universities and drug laboratories (some international).

Role
Participant
Institution
Dom Bosco Catholic University
Years
2005 - 2007