Active vision of bees in a simple pattern discrimination task
- School of Biological and Behavioural Sciences, Queen Mary University of London, United Kingdom
- Department of Computer Science, University of Sheffield, United Kingdom
- School of Biosciences, University of Sheffield, United Kingdom
- School of Natural Sciences, Macquarie University, Australia
- Drone Development Lab, Ben Thorns Ltd, United Kingdom
Figures
Figure 1 with 1 supplement
Bees’ performance in a pattern recognition task.
(A) Illustration of the flight arena, equipped with two cameras: a side camera above the entrance viewing the rear wall and a top camera above the rear wall recording from above. (B) Training and testing protocol: Bees were trained with four plus and four multiplication patterns on white discs (10 cm diameter, 2 mm margin), attached to the arena’s rear wall via small Eppendorf tubes. Group 1 of bumblebees (n=20) was rewarded with the plus sign (sucrose solution) and punished with the multiplication sign (quinine), while Group 2 (n=20) received the opposite reinforcement. After training, four tests with refresher interval bouts were conducted: (1) a learning test using the original patterns from training, (2) a top-half test showing only the upper half of the patterns, (3) a bottom-half test displaying only the lower half of the patterns, and (4) a single-bar test presenting individual bars in different orientations. (C) Learning curves for Group 1 (blue) and Group 2 (orange) show both groups learned the pattern discrimination task. (D) Unrewarded learning test performance, with both groups distinguishing patterns (p<4.8e-3). (E) Initial inspections of the plus symbol upon arena entry showed no significant preference from a distance (>5 cm), with 8 out of 13 bees in group 1 and 5 out of 12 bees in group 2 making initial visits to the plus symbol in the learning test (p=0.55).
Figure 1—figure supplement 1
Bees’ flight paths upon first entering the arena during learning tests.
In 10 instances the bees initially inspected the correct stimulus, scanned the pattern and visited the feeder. In the remaining nine flights the bees initially inspected the incorrect pattern, then rejected the pattern and flew to another, usually adjacent pattern. One video is missing where the footage was only recorded at 30 fps; this bee initially inspected the incorrect pattern, and again rejected the stimulus. The bees’ first inspection appears to be random with 50/50 correct pattern selections from the arena entrance; this suggests bees have to scan the stimuli before making decisions. Line colour from blue to yellow: flight speed 0.0–0.7 m/s.
Figure 2 with 1 supplement
Bumblebees' performance in the transfer testing phase.
This figure shows bumblebee performance in unrewarded tests where different sections of the training patterns were presented. (A) Bees successfully discriminated the bottom halves of the patterns, with statistically significant results (p < 7.7e-3). However, in (B), bees exposed only to the top halves did not show significant recognition of the correct patterns (p=0.83), suggesting these cues were insufficient for accurate discrimination. (C) shows that bees exhibited a significantly higher response to individual bars matching the rewarded orientations in the training (p<6.8e-4), indicating effective recognition based on specific visual cues learned during training.
Figure 2—figure supplement 1
Bumblebees' preference in the single-bar orientation test.
The bar plot shows the percentage of choices made by bumblebees in the single-bar test, highlighting bees' differential responses to isolated bar cues. Bees trained to the plus pattern displayed a strong preference for the vertical bar, whereas bees trained to the multiplication pattern showed significantly higher selection of both the 45° and –45° bars (p<3.0e-4), with no notable difference in their choices between these two orientations.
Figure 3
Scanning behaviour of bumblebees in a simple pattern discrimination task (A, B).
An example of a flight path showing the activity of a bee during part of the learning test; presented bee trained to select the 'plus' and avoid the 'multiplication' patterns. Each point on the flight path corresponds to a single video frame with an interval of 4ms between frames which was recorded from the front camera (A) and top camera (B). The bee sequentially inspected each pattern, correctly landed on the multiplication sign and avoided the plus sign. The colour map changes from blue to yellow with increasing time (See Video 1). In (A) the plus patterns may appear slightly misaligned vertically due to visual distortion caused by the camera perspective and edge warping. In the actual experimental setup, the patterns were presented to the bees with a precise vertical alignment. In (B) the black lines in the right panel exhibit bee’s body yaw orientations. (C) Distribution of average entrance flight speed toward the wall in the learning tests (See Figure 1). Filled dots: speed of each individual bee. (D) Probability distribution of the bees’ speed in two conditions; when they were inspecting patterns (Red), and when they were flying between patterns (Yellow). This indicates that they slow down to scan patterns before accepting while they accelerate when they fly to another pattern. (E) Probability distribution of the bees’ distance from the stimuli wall whilst inspecting patterns. (F) From video analysis, the proportion of scanned regions (mean ± SEM) of bees’ inspections before the correct accept or correct rejection (x-axis: regions of interest are highlighted in grey). Triangles: inspection proportion (mean ± SEM) of Group 1 bees (trained to plus sign rewarding), squares: Group 2 bees (trained to multiplication sign), green error bars: correct accepts; red error bars: correct rejections.
Figure 4
Bee scanning strategy in a pattern recognition task.
(A) The flight paths of one example of acceptance and two examples of rejection behaviours of a bee trained to plus; the bee accepted the plus pattern after scanning the lower half of the vertical bar, while she rejected the multiplication pattern after scanning one or both diagonal bars. Line colour: flight speed 0.0–0.4 ms–1 (see Video 2). (B) Group 1 (trained to plus) probability maps (heat-maps) of bees’ locations per frame in front of plus and multiplication type stimuli during all learning tests. The yellow colours show the most visited regions. (C & D) same analysis as A, B for Group 2 bees trained to discriminate the multiplication sign from the plus sign. This indicates that bees typically scanned the lower half of the pattern with a lower speed to Group 1 bees, prior to their decisions (see Videos 3 and 4). (E) Three examples of bees’ flight paths shown from the top camera; black lines show bees’ body orientation during the flight, and arrows designate the start and ending time of scanning. (F) Probability distribution of the bees’ body yaw orientation perpendicular to the rear stimuli wall in two conditions: when they were inspecting patterns (red) and when they were flying between patterns (yellow). Inset figure exhibits one example of a bee’s orientation (viewed from the top) with +45° to the orientation of the wall on which stimuli were displayed. Bees viewed the patterns at a median ~±30° whilst scanning, with one or other eye having a predominant view. On the other hand, when they transitioned between patterns, the body orientation was more parallel to the flight direction with a wider distribution of orientations relative to the stimuli wall, resulting in a median of ~±50° to the stimuli wall. The dashed lines show the Gaussian mixture distribution models were fitted to each distribution (flights within patterns:; flights between patterns:). (G) Mean flight speed (±SEM) of scanning flight prior to decisions (accept and rejection) for both groups of bees. Blue: Group 1 (trained to plus sign); orange: Group 2 (trained to multiplication sign). (H) Inspection time (i.e., the time spent hovering in front of a pattern) for each symbol type for both groups of bees; inspection times of bees in front of both pattern types were equal regardless of their decision or training protocol.
Figure 5
Bumblebee scanning behaviour when inspecting partial and component visual cues.
(A & B) Heatmaps of bumblebee flight paths overlaid on the visual stimuli during the bottom-half test (A) and top-half test (B). Each panel is grouped based on the bees' decisions to correctly or incorrectly accept or reject the presented test patterns. The heatmaps illustrate areas where bees spent the most time hovering, with warmer colours indicating prolonged hovering and increased attention to specific regions of the pattern. The contour lines further highlight the spatial distribution of bee activity over the stimuli. (C) presents bar plots comparing hovering times for correct accept and correct reject responses. The results reveal a strong focus on the bottom sections with longer hovering time when bees correctly accepted the positive pattern (p<2.8e-4), suggesting a scanning strategy that prioritises these areas for pattern discrimination. However, there was no significant difference in hovering time when bees correctly rejected the negative patterns (p>0.05). This preferential focus on specific regions implies that bees adopt scanning behaviours centred on discriminative visual cues, particularly in the lower half of the patterns, as an effective strategy for pattern recognition.
Videos
Video 1
Example video of a bee’s scanning behaviour in the learning test.
The bee was trained to find the reward from the multiplication. The bee inspected lower region of the patterns and rejected the plus and accepted several multiplication signs. The video was recorded by the front camera at 240 fps (frames per second).
Video 2
Example video of rejecting the multiplication and accept the plus.
The video was recorded at 240 fps (frames per second).
Video 3
Example video of rejecting two plus signs and accept the multiplication sign.
The video was recorded at 240 fps (frames per second).
Video 4
Example video of from the top camera.
The video is the same flight track as Video 3 but recorded from the top camera. The video was recorded at 120 fps (frames per second).
Tables
Table 1
Summary of the Generalised Linear Mixed Model (GLMM) examining factors in relation to proportion of correct choices during the training.
The dependent variable was the number of correct choices from the block of 10 choices. Fixed factors, Colony, Group and Trial were examined in the model. Bee index was calculated in the model as a random factor. Model fit statistics: AIC = 32.91; BIC = 337.62; Log-Likelihood=−156.45; Deviance = 312.91.
| Fixed factors | Estimate | SE | tStat | DF | p Value | Lower | Upper |
|---|---|---|---|---|---|---|---|
| Intercept | –0.11 | 0.50 | –0.33 | 266 | 0.71 | –1.17 | 0.83 |
| Colony | 0.19 | 0.17 | 1.04 | 266 | 0.25 | –0.13 | 0.42 |
| Group | –0.06 | 0.27 | –0.24 | 266 | 0.82 | –0.61 | 0.36 |
| Trials | 0.21 | 0.03 | 6.75 | 266 | 4.8e-9 | 0.14 | 0.27 |
Additional files
Download links
A two-part list of links to download the article, or parts of the article, in various formats.
Downloads (link to download the article as PDF)
Open citations (links to open the citations from this article in various online reference manager services)
Cite this article (links to download the citations from this article in formats compatible with various reference manager tools)
Active vision of bees in a simple pattern discrimination task
eLife 14:e106332.
https://doi.org/10.7554/eLife.106332
