Multiple contributions at SIDIM 2019

Our latest work was presented at SIDIM 2019 (Seminario Interuniversitario de Investigación en Ciencias Matemáticas), at UPR Humacao, Puerto Rico

Presentation on the integration of detection, tracking and pollen detection for the automatic system that can characterize foraging patterns and differentiate between pollen bearing and non-pollen bearing bees. First results on 8 days of monitoring were shown.
Ivan F. Rodriguez, Kristin Branson, Edgar Acuña, Rémi Mégret, José L. Agosto-Rivera, Tugrul Giray, “Automatic monitoring of the foraging behavior of tagged and untagged honey bees”, SIDIM 2019, Humacao, March 1-2 2019.

Presentation on how to leverage automatic tracking and non-supervised learning to train neural networks to recognize the identity of a bee. The technique reached 0.94 ROC AUC for deciding whether two images of bees were the same bee or not, based on extensive video collection.
Jeffrey Chan, Ivan F. Rodriguez, Rémi Mégret, José L. Agosto-Rivera, Tugrul Giray, “Learning good features to discriminate untagged bees in video using non- supervised learning”, SIDIM 2019, Humacao, March 1-2 2019.

Poster on the detection of fanning behavior. Joint work with Edward Latorre and his students from UPR Arecibo, thanks to support from PR-LSAMP Faculty Summer Research Program.
Edward Latorre, Kelvin López, Ivan F. Rodríguez, Matías Rosner Ortiz, Rémi Mégret, Tugrul Giray, José L. Agosto, “Recognition of Fanning Bees from Video using Convolutional Neural Networks”, SIDIM 2019, Humacao, March 1-2 2019.

Paper: Clustering Honeybees by its Daily Activity

Edgar Acuña presented our paper about the analysis of RFID data provided by the team of Yves Le Conte at INRIA Avignon, France. Clustering was applied to determine bees with similar activity and to estimate the time during the day when the bees are most active:

Edgar Acuna, Velcy Palomino, Jose Agosto, Rémi Mégret, Tugrul Giray, Alberto Prado, Cédric Alaux and Yves Le Conte, “Clustering Honeybees by its Daily Activity”, International Conference on Pattern Recognition Applications and Methods (ICPRAM), Prague, Czech Republic, Feb 19-21 2019.

Paper: Multiple Insect and Animal Tracking in Video using Part Affinity Fields

Master student Ivan Rodriguez presented his work at ICPR conference in Beijin:

Ivan F. Rodríguez, Rémi Mégret, Roian Egnor, Kristin Branson, José L. Agosto, Tugrul Giray, Edgar Acuña . “Multiple Insect and Animal Tracking in Video using Part Affinity Fields”, Visual observation and analysis of Vertebrate And Insect Behavior (VAIB), International Conference on Pattern Recognition (ICPR), Beijin, China, 20-24 August 2018. [PDF]

Pose estimation consists in detecting the parts (head, thorax, abdomen, and the two antennas) of all honeybees automatically, and connecting them to produce a “skeleton” for each individual. Using a Deep Convolutional Network, all bees in the same image are processed jointly without the need of detecting the bees in advance.

Pose estimation for multiple honeybees at the entrance of the hive.

Tracking then produces trajectories that can be classified at entering/exiting automatically.

Trajectories at the entrance. E=entering, L=leaving.

Paper: Applying Functional Data Clustering for Analyzing Circadian Cycles of Honeybees

New paper was presented by Edgar Acuña this summer:
R. Trespalacios, E. Acuña, V. Palomino, J. Agosto, R. Mégret, M. A. Giannoni-Guzman, “Applying Functional Data Clustering for Analyzing Circadian Cycles of Honeybees”, International Conference on Data Mining (ICDM), Istanbul, Turkey, July 2018.

Abstract: In this paper, we analyze the periodic cycle of honey- bees when they have between 7 and 9 days of age. The circadian clock of the bees present very erratic behavior that it is a challenge to detect cycles. In signal processing, there are several methods to detect periodic patterns. In here, we will use a well-known test, named periodogram, to evaluate rhythmicity and estimate the period. Besides, to determine whether or no rhythmicity exists, we estimate the time when the bees behavior starts to be rhythmic. Also, it can occur that the bees behavior never gets rhythmic. We perform consecutive test of rhythmicity until find out periodicity, if this exists. The instant in which the time series becomes periodic is considered the moment in which the bees activity starts. Furthermore, we carry out the periodicity test for the time series obtained from the actogram. We find out that for bees which time series is visually periodic, our method detects correctly the starting time. However, for bees which time series does not show a cyclic pattern our method fails due to a very erratic time series and that the consecutive test results also will show this erratic behavior. Finally, we classify the bees according to theirs beginning of a periodic cycle, using functional data analysis.

R package CircadianDynamics

This R-package contains various functions to clean, visualize and analyze the dynamics of circadian parameters from time series of several types of data. The types of data and formats supported in the package include: 1) temperature, humidity, and light from the HOBO sensors (U12-012); 2) barometric pressure, wind speed and direction from the HOBO Micro weather station (H21-USB); 3) colony weight from the Brushy Mountain Hive Scales (714 – Hive Scale); 4) locomotor activity from all DAM and LAM series of Activity Monitors from Trikinetics ( and; 4) time stamp annotated events from videos of the hive entrance. The unique feature of this package compared to existing tools, is the ability to quantify the dynamics of circadian parameters over time rather than assuming static rhythms. In general, this package estimates the period, amplitude, phase and rhythm strength of a sliding window from the time series inputted using 3 different methods that are well establish in the circadian field (references). In addition, functions to upload and fix common issues such as date, temporal formats, gaps and changes in sampling rate from each of these types of files are also included in the package.

Credits: Jonathan Aleman Rios, Jose L. Agosto-Rivera, Rémi Mégret, Tugrul Giray, Edgar Acuña.
License: BSD

Download: CircadianDynamics_scripts

Short paper RFIAP2018

Our short paper was accepted at RFIAP. This work explore the use of Convolutional Neural Networks for the pose estimation of honeybees (localization of their body parts) from images, using a user trainable model. It shows promising results with over 95% of correct detection. These results open new possibilities for automated tracking of a rich set of honeybee behavior at their colony in the field.

[1] I. F. Rodríguez, K. Branson, E. Acuña, J. L. Agosto-Rivera, T. Giray, R. Mégret, “Honeybee Detection and Pose Estimation using Convolutional Neural Networks”, Reconnaissance des Formes, Image, Apprentissage et Perception, Paris, june 2018. [PDF]

Meeting May 18

Next project meeting will be held in UPR Río Piedras, College of Natural Science, Friday May 18. 2018.

Participants will meet at 9am in room A-233.
General public is welcome to come discuss during the poster session, 1-3pm in the vestibule (see flyer: BigDBee-flyer-poster-session-2018May18)

PRISM 2018: Oral presentations

Congratulation to Christian Esteves, Emmanuel Nieves and Jeffrey Chan for their oral presentation at PRLSAMP’s Puerto Rico Interdisciplinary Scientific Meeting (PRISM) 2018, as part of the UPRRP Math and CCOM delegation !

The abstracts can be downloaded here:

[1] Christian Esteves, Emmanuel Nieves, Rëmi Mégret, Iván Rodriguez: “Web App development for multi user annotation interface”, Puerto Rico Interdisciplinary Scientific Meeting (PRISM), April 2018, Gurabo, PR
[2] Jeffrey Chan, Iván Rodriguez, Rémi Mégret, Edgar Acuña, José. L. Agosto, Tugrul Giray: “Extracting Fingerprints for Bee Identification”, Puerto Rico Interdisciplinary Scientific Meeting (PRISM), April 2018, Gurabo, PR

UPRRP Math and Computer Science delegation at PRISM 2018

Open positions: research assistants

Open positions for undergraduate and graduate research assistants in Computer Science and Data Science

Send us an email, join the team !

The project provides opportunities to work on various topics, according to the skills and motivation of the students interested:

– Computer vision: automatic analysis of video data (bee tracking, activity recognition…)

– Webapp development: app to crowd-source data analysis, client and server side components

– Scientific visualization: display of bees’ activities, interactive exploration of data

– High Performance Computing: deployment of computations on GPU node equipped with state-of-the-art NVIDIA Tesla cards and on linux clusters…

Data Mining: explore and identify patterns in the long-term behavior of the bees

Download the Flyer