Skip to main content

Research

Detection, Tracking and Recognition of Agricultural Pests Using Hyperspectral Imaging

Keywords : Hyperspectral imaging · Butterfly recognition · Multi-target tracking · Pest insects · Hyperspectral video · Precision agriculture

This research activity corresponds to my PhD work, focusing on the design of a complete pipeline for detecting, tracking and recognizing moving agricultural pests from hyperspectral imagery, with applications to crop monitoring.

Research context

My PhD was conducted under a joint supervision agreement between the University of Abomey-Calavi and the Université du Littoral Côte d'Opale, within the Doctoral Schools of Engineering Sciences and Sciences, Technology and Health.

The research was carried out jointly at the LISIC laboratory and the LRSIA laboratory, with co-funding from ULCO and the International Institute of Tropical Agriculture (IITA, Benin).

Research activities were distributed between the two laboratories according to their respective expertise, fostering strong scientific synergy between data acquisition, spectral recognition and insect tracking.

The thesis, supervised by Arnaud Ahouandjinou, Gilles Delmaire and Gilles Roussel, was defended on October 23, 2025; the manuscript is available on HAL: tel-05371801.

Work description

Traditional pest control strategies are often reactive and lead to excessive pesticide use. A more sustainable alternative relies on early detection of pests at the adult stage.

The main objective of this PhD is to design hyperspectral-based methods for the detection, tracking and recognition of moving pests by jointly exploiting spatial, spectral and temporal information.

The work relies on a compact spatio-spectral camera, requiring joint modeling of spectral signatures and insect flight dynamics.

Contributions

Data acquisition and preparation

Two complementary acquisition systems were developed: a low-cost controlled-motion setup (about 100 euros, motorized conveyor belt) for immobile insects, used at IITA to reconstruct reference hyperspectral cubes, and an autonomous monitoring system deployed in real conditions at the Beau Pays botanical garden in Marck, combining a spatio-spectral camera and an NVIDIA Jetson board. Running at 25 frames per second, the latter only keeps sequences showing activity, detected by background subtraction.

Spectral species recognition

Proposal of a Bayesian approach based on a convex combination of Gaussian Naive Bayes and Z-score, with four classification strategies, for butterfly species recognition from raw spatio-spectral image sequences: 88.75% accuracy with a single image and 97.5% when aggregating four images, with four images processed in 741 ms on a standard CPU.

Multi-target tracking based on flight dynamics

Development of a multi-target tracking method combining Kalman filters with a flight deformation model (wing deformations, orientation variations): the extended Kalman filter reaches 73% IoU with no identity switch over 15 sequences, compared with 62% and 3 switches for the linear filter.

Identification and monitoring pipeline

Design of a complete pipeline for identifying pest butterflies in the wild and estimating their density from spatio-spectral images: to our knowledge, the first hyperspectral approach to pest insect detection under real-world conditions and in near real time (publication in preparation).

  1. [R1]

    Statistical framework for butterfly species recognition using raw spatio-hyperspectral images

    E. Adjé, G. Delmaire, A.S.R.M. Ahouandjinou, M. Puigt, G. Roussel

    Pattern Analysis and Applications, vol. 28, n° 3, pp. 136, 2025

  2. [CI3]

    Deformation-Aware Butterfly Tracking in Raw Spatio-Spectral Images

    E. Adjé, A.S.R.M. Ahouandjinou, G. Delmaire, P. Kiki, G. Roussel

    Pattern Recognition and Image Analysis (IbPRIA 2025), vol. 15937, pp. 41–53, 2025

  3. [CN1]

    Suivi de papillons basé sur les déformations du mouvement de vol à partir d'images spatio-spectrales

    E. Adjé, A.S.R.M. Ahouandjinou, G. Delmaire, P. Kiki, G. Roussel

    XXXe Colloque Francophone de Traitement du Signal et des Images (GRETSI 2025), pp. 417–420, 2025

  4. [T1]

    Détection, suivi et reconnaissance d'insectes en mouvement par imagerie hyperspectrale : application à la surveillance des ravageurs en agriculture

    E. Adjé

    Université du Littoral Côte d'Opale & Université d'Abomey-Calavi, 2025

  5. [CI1]

    Towards Automatic Butterfly Species Recognition Using a Single Spatio-Hyperspectral Image

    E. Adjé, G. Delmaire, A.S.R.M. Ahouandjinou, M. Puigt, G. Roussel

    32nd European Signal Processing Conference (EUSIPCO 2024), pp. 1252–1256, 2024

  6. [CI2]

    Advancements in Video-Based Insect Tracking: A Bibliometric Analysis to a Short Survey

    E. Adjé, A.S.R.M. Ahouandjinou, G. Delmaire, G. Roussel, V.R. Houndji

    2023 7th International Conference on Advances in Image Processing (ICAIP), pp. 75–82, 2024

  7. [CI4]

    Robust Approach for Butterfly Species Classification Using a Single Spatio-Hyperspectral Image

    E. Adjé, G. Delmaire, A.S.R.M. Ahouandjinou, M. Puigt, G. Roussel

    14th Workshop on Hyperspectral Imaging and Signal Processing: Evolution in Remote Sensing (WHISPERS 2024), pp. 1–5, 2024

See full list of publications