BeeNet: Reconstructing Flower Shapes from Electric Fields using Deep Learning

TL;DR

This work introduces a deep learning framework for solving the inverse electrostatic imaging problem, enabling object shape reconstruction directly from measured electric fields, and develops an algorithm that infers the shapes of polarisable flowers from the electric field generated in response to a nearby charged arthropod.

Abstract

Pollinating insects can obtain information from electric fields arising from flowers. The density and usefulness of electric information remain unknown. Here, we show that electric information can be used to reconstruct geometrical features of the field source. We develop an algorithm that infers the shapes of polarisable flowers from the electric field generated in response to a nearby charged arthropod. We computed the electric fields arising from arthropod flower interactions for varying petal geometries, and used these data to train a deep learning U Net model to recreate the floral shapes. The model accurately reconstructed diverse shapes, including more complex flower morphologies not included in training. Reconstruction performance peaked at an optimal arthropod flower distance, indicating distance dependent encoding of shape information. These findings indicate that electroreception can impart rich spatial detail, offering insights into the electric ecology of arthropods. Together, this work introduces a deep learning framework for solving the inverse electrostatic imaging problem, enabling object shape reconstruction directly from measured electric fields.

Figures (4)

• Figure 1: A) Diagram of an arthropod and flower showing how their interaction alters the electric fields. B) The three resulting fields rendered as images: red and green represent perturbation fields in the x and y directions, and blue represents the electric potential. C) These three channels merged into a single RGB image. D) A simplified schematic of the BeeNet model based on standard U-Net architecture. E) Model predictions alongside the corresponding ground truth. F) The various flower configurations used during training.
• Figure 2: A) The output of the model overlaying the ground truth. White pixels are correctly predicted (True positive). Green and magenta are the false negatives and false positives, respectively. B-E) The ground truth with the overlapping images displaying the accuracy for each of the shape perturbations
• Figure 3: A) Violin plot of the F1 score for each number of petals in flowers. The valuation dataset was used for flowers with 1-3 petals (blue) and test set was used for 4-petal flowers (orange). B) F1 score as a function of petal permeability, evaluated on the validation dataset. C) F1 score as a function of arthropod–flower distance, evaluated on the validation dataset.
• Figure 4: A) Overlapping images displaying the accuracy for different rotations of the 4 petal flowers. B) 4 petal flowers with different levels of pointedness. (Green and magenta are the false negatives and false positives, respectively.)