Vision of the honeybee Apis mellifera for patterns with one pair of equal orthogonal bars

The visual discrimination of patterns of two equal orthogonal black bars by honeybees has been studied in a Y-choice apparatus with the patterns presented vertically at a fixed range. Previous work shows that bees can discriminate the locations of one, or possibly more, contrasts in targets that are in the same position throughout the training. Therefore, in critical experiments, the locations of areas of black were regularly shuffled to make them useless as cues. The bees discriminate consistent radial and tangential cues irrespective of their location on the target during learning and testing. Orientation cues, to be discriminated, must be presented on corresponding sides of the two targets. When orientation, radial and tangential cues are omitted or made useless by alternating them, discrimination is impossible, although the patterns may look quite different to us. The shape or the layout of local cues is not re-assembled from the locations of the bars, even when there are only two bars in the pattern, as if the bees cannot locate the individual bars within the large spatial fields of their global filters.     

Tactile learning in the honeybee

Free-flying bees were conditioned on a vertical wall to a vertical tactile pattern consisting of parallel lines of grooves and elevations. The asymptote of the learning curve is reached after approximately 25 rewards. Bees can discriminate the conditioned vertical pattern from a horizontal or diagonal alternative. Angle discrimination is apparent only for relatively coarse tactile cues. The proboscis extension response of fixed bees was used to condition bees to a vertical tactile pattern which was presented to the antennae. The learning curve reaches an asymptote after 4 rewards. After 7 unrewarded extinction trials the conditioned responses are reduced to 50%. Bees show best discrimination for patterns whose edges they can scan with their antennae. The animals show a high degree of generalization by responding to an object irrespective of the trained pattern. Under laboratory conditions fixed bees can discriminate the angles and spatial wavelengths of fine tactile patterns consisting of parallel grooves. Bees can also discriminate forms and sizes of tactile patterns. They do not discriminate between different types of edges and between positive and negative forms. Accepted: 17 September 1998     

Visual resolution of the orientation cue by the honeybee (Apis mellifera)

Bees were trained to discriminate between a pattern with two or more black bars and a similar pattern with the bars at right angles. Earlier measures of the resolution of oblique black and white regular gratings of different periods were confirmed. The positions of the training bars were shifted every 5 min to prevent the bees from using their locations as cues. To measure the length of the detectors of edge orientation, the trained bees were tested with targets filled with parallel short black/white edges of various lengths. The minimum individual length of edge required to discriminate the orientation cue was found to be near 3 degrees, and similar for vertical, horizontal and oblique edges. This is the first time that this kind of resolution has been measured in an invertebrate. The bees learn and recognize the edge orientation, not the lay-out of the pattern.     

Pattern discrimination by the honeybee (Apis mellifera ) is colour blind for radial / tangential cues

Bees were trained to discriminate between two patterns, one of which was associated with a reward, in a Y-choice apparatus with the targets presented vertically at a distance at an angular subtense of 50°. Previous work with this apparatus has found discrimination between two patterns of coloured gratings or radial sectors that are fixed in different orientations during the training. When there was contrast to the blue receptors alone, gratings of period 6° were resolved, and 4° when there was contrast to the green receptors. In the present work, bees discriminate between a pattern containing tangentially arranged edges and one containing radially arranged edges, both with no average edge orientation. The targets were rotated every 5 min to make the locations of areas useless as cues. The edges remained consistently radial or tangential and were therefore the only cues. Tests with patterns of selected colours and various levels of grey show that for each colour there is a level of grey at which discrimination fails. Discrimination is therefore colour-blind. The same patterns were made with combinations of coloured papers that give no contrast to the green receptors or alternatively to the blue receptors. The bees discriminate only if the edges between colours present a contrast to the green receptors. The system that discriminates generalized radial and tangential cues is therefore colour blind because the inputs are restricted to the green receptors, not because receptor outputs are added together. The same result was obtained with a very coarse pattern of period 20°. Accepted: 10 January 1999     

Movement learning of free flying honeybees

Summary Free flying honeybees were conditioned to moving black and white stripe patterns. Bees learn rapidly to distinguish the direction of movement in the vertical and horizontal plane.After being trained to a moving pattern bees do not discriminate the moving alternative from a stationary one. There is no significant velocity discrimination for patterns moving in the same direction.For vertical movements there are clear asymmetries in the spontaneous choice preference and in the learning curves for patterns moving upward or downward.After bees are trained to a stationary pattern they can discriminate it from an upward moving alternative. Learning curves involving movement are generally biphasic, suggesting different adaptive systems depending on the number of rewards.The flight pattern of bees which are trained to movement changes during the process of learning. At the beginning of the learning procedure bees reveal an optokinetic response to the moving patterns, this response is strongly reduced after a number of rewards on a moving pattern.     

Pattern recognition in honeybees: eidetic imagery and orientation discrimination

The roles of eidetic imagery and orientational cues, respectively, in the discrimination of visual patterns by honeybees (Apis mellifera) were evaluated by training the bees to discriminate between patterns consisting of periodic, black and white square wave gratings. Training and tests with a number of different pairs of patterns revealed that bees use orientational cues almost exclusively, if such are present, and make use of eidetic images only when orientational cues are not available. On the other hand, if a pattern carries strong orientational cues, bees learn the orientation even if it is irrelevant to the discrimination task on which they are trained.     

Pattern vision of the honeybee (Apis mellifera). What is an oriented edge?

Pairs of black patterns on a white background, one rewarded the other not, were presented vertically each in one arm of a Y-maze. During training the locations of the black areas were changed every 5 min to prevent the bees using them as cues, but cues from edges were kept consistent. Bees detect orientation even in a gradient that subtends 36° from black to white (normal to the edge). Orientation cues in short lengths of edge are detected and summed on each side of the fixation point, irrespective of the lay-out of the pattern. Edges at right angles reduce the total orientation cue. The polarity of edges in a sawtooth grating is weakly discriminated, but not the orientation of a fault line where two gratings meet. Edge quality can be discriminated, but is not recognised in unfamiliar orientations. When spot location is excluded as a cue, the orientation of a row of spots or squares which individually provide no net orientation cue is not discriminated. In conclusion, when locations of black areas are shuffled, the bees remember the sum of local orientation cues but not the global pattern, and there is no re-assembly of a pattern based on differently oriented edges. A neuronal model consistent with these results is presented. Accepted: 5 March 2000     

Discrimination of coloured patterns by honeybees through chromatic and achromatic cues

We investigated pattern discrimination by worker honeybees, Apis mellifera, focusing on the roles of spectral cues and the angular size of patterns. Free-flying bees were trained to discriminate concentric patterns in a Y-maze. The rewarded pattern could be composed of either a cyan and a yellow colour, which presented both different chromatic and achromatic L-receptor contrast, or an orange and a blue colour, which presented different chromatic cues, but the same L-receptor contrast. The non-rewarded alternative was either a single-coloured disc with the colour of the central disc or the surrounding ring of the pattern, a checkerboard pattern with non-resolvable squares, the reversed pattern, or the elements of the training pattern (disc or ring alone). Bees resolved and learned both colour elements in the rewarded patterns and their spatial properties. When the patterns subtended large visual angles, this discrimination used chromatic cues only. Patterns with yellow or orange central discs were generalised toward the yellow and orange colours, respectively. When the patterns subtended a visual angle close to the detection limit and L-receptor contrast was mediating discrimination, pattern perception was reduced: bees perceived only the pattern element with higher contrast.     

Generalization in visual recognition by the honeybee (Apis mellifera) : A review and explanation

During a century of studies on honeybee vision, generalization was the word for the acceptance of an unfamiliar pattern in the place of the training pattern, or the ability to learn a common factor in a group of related patterns. The ideas that bees generalize one pattern for another, detect similarity and differences, or form categories, were derived from the use of the same terms in the human cognitive sciences. Recent work now reveals a mechanistic explanation for bees. Small groups of ommatidia converge upon feature detectors that respond selectively to certain parameters that are in the pattern: modulation in the receptors, edge orientations, or to areas of black or colour. Within each local region of the eye the responses of each type of feature detector are summed to form a cue. The cues are therefore not in the pattern, but are local totals in the bee. Each cue has a quality, a quantity and a position on the eye, like a neuron response. This summation of edge detector responses destroys the local pattern based on edge orientation but preserves a coarse, sparse and simplified version of the panorama. In order of preference, the cues are: local receptor modulation, positions of well-separated black areas, a small black spot, colour and positions of the centres of each cue, radial edges, the averaged edge orientation and tangential edges. A pattern is always accepted by a trained bee that detects the expected cues in the expected places and no unexpected cues. The actual patterns are irrelevant. Therefore we have an explanation of generalization that is based on experimental testing of trained bees, not by analogy with other animals.Historically, generalization appeared when the training patterns were regularly interchanged to make the bees examine them. This strategy forced the bees to ignore parameters outside the training pattern, so that learning was restricted to one local eye region. This in turn limited the memory to one cue of each type, so that recognition was ambiguous because the cues were insufficient to distinguish all patterns. On the other hand, bees trained on very large targets, or by landing on the pattern, learned cues in several eye regions, and were able to recognize the coarse configural layout.     

Some labels that are recognized on landmarks by the honeybee (Apis mellifera)

Freely flying bees were trained in a situation that resembled the natural task of a bee arriving at a foraging site that was located by a landmark. The bees' task was to locate the reward in the arm of the Y-choice apparatus, where a black pattern on a white background was displayed in one arm versus a white target in the other arm, at a range of 27 cm. The alternative patterns for the training included previously identified cues. They were: an oblique bar, three parallel oblique bars, an oblique grating, a square cross, six spokes, a large or a small spot, a spotty modulation, or a ring. The trained bees were given a variety of interleaved tests to discover the labels they had used to identify the patterns. A label is defined as the coincidence of cues that contributed to the recognition of a single landmark. The bees learned, firstly, the black area at the expected place, secondly, modulation caused by edges at the expected place. These cues were quantified and always available. In addition, the orientation cue was learned from a grating that covered the target, but was ignored in a single bar. The bees learned the positions of the centres of black and of radial symmetry. In tests, they also recognized unfamiliar cues that were not displayed in the training. The cues and preferences were similar to those used to discriminate between two targets. The new experiments validate some old conclusions that have been controversial for 40 years.     

The preferences of the honeybee (Apis mellifera) for different visual cues during the learning process

By working with very simple images, a number of different visual cues used by the honeybee have been described over the past decades. In most of the work, the bees had no control over the choice of the images, and it was not clear whether they learned the rewarded pattern or the difference between two images. Preferences were known to exist when untrained bees selected one pattern from a variety of them, but because the preferences of the bees were ignored, it was not possible to understand how natural images displaying several cues were detected. The preferences were also essential to make a computer model of the visual system. Therefore experiments were devised to show the order of preference for the known cues in the training situation. Freely flying bees were trained to discriminate between a rewarded target with one pattern on the left side and a different one on the right, versus a white or neutral target. This arrangement gave the bees a choice of what to learn. Tests showed that in some cases they learned two or three cues simultaneously; in other cases the bees learned one, or they preferred to avoid the unrewarded target. By testing with different combinations of patterns, it was possible to put the cues into an order of preference. Of the known cues, loosely or tightly attached to eye coordinates, a black or blue spot was the most preferred, followed by strong modulation caused by edges, the orientation of parallel bars, six equally spaced spokes, a clean white target, and then a square cross and a ring. A patch of blue colour was preferred to yellow.     

Biological significance of distinguishing between similar colours in spectrally variable illumination: bumblebees (<Emphasis Type="Italic">Bombus terrestris</Emphasis>) as a case study

Individual bumblebees were trained to choose between rewarded target flowers and non-rewarded distractor flowers in a controlled illumination laboratory. Bees learnt to discriminate similar colours, but with smaller colour distances the frequency of errors increased. This indicates that pollen transfer might occur between flowers with similar colours, even if these colours are distinguishable. The effect of similar colours on reducing foraging accuracy of bees is evident for colour distances high above discrimination threshold, which explains previous field observations showing that bees do not exhibit complete flower constancy unless flower colour between species is distinct. Bees tested in spectrally different illumination conditions experienced a significant decrease in their ability to discriminate between similar colours. The extent to which this happens differs in different areas of colour space, which is consistent with a von Kries-type model of colour constancy. We find that it would be beneficial for plant species to have highly distinctive colour signals to overcome limitations on the bees performance in reliably judging differences between similar colours. An exception to this finding was flowers that varied in shape, in which case bees used this cue to compensate for inaccuracies of colour vision.     

What the honeybee sees: a review of the recognition system of Apis mellifera

Abstract. For many years, two opposing theories have dominated our ideas of what honeybees see. The earliest proposal based on training experiments was that bees detected only simple attributes or features, irrespective of the actual pattern. The features demonstrated experimentally before 1940 were the disruption of the pattern (related to spatial frequency), the area of black or colour, the length of edge, and the angle of orientation of a bar or grating. Cues discovered recently are the range, and radial and tangential edges, and symmetry, relative to the fixation point, which is usually the reward hole. This theory could not explain why recognition failed when the pattern was moved. In the second theory, proposed in 1969, the bee detected the retinotopic directions of black or coloured areas, and estimated the areas of overlap and nonoverlap on each test pattern with the corresponding positions in the training pattern. This proposal explained the progressive loss of recognition as a test pattern was moved or reduced in size, but required that the bees saw and remembered the layout of every learned pattern and calculated the mismatch with each test image. Even so, the same measure of the mismatch was given by many test patterns and could not detect a pattern uniquely. Moreover, this theory could not explain the abundant evidence of simple feature detectors. Recent work has shown that bees learn one or more of a limited number of simple cues. A newly discovered cue is the position, mainly in the vertical direction, of the common centre (centroid) of black areas combined together. Significantly, however, the trained bees look for the cues mentioned above only in the range of places where they had occurred during the training. These two observations made possible a synthesis of both theories. There is no experimental evidence that the bees detect or re-assemble the layout of patterns in space; instead, they look for a cue in the expected place. With an array of detectors of the known cues, together with their directions, this mechanism would enable bees to recognize each familiar place from the coincidences of cues in different directions around the head.     

Bee vision of pattern and 3D. The Bidder Lecture 1994

Insect vision is nothing if not active. The regular head movements, called saccades, enable the fly Drosophila to keep a straight path in flight despite inequalities in the thrust of the wings. Using their own motion, bees in flight measure the ranges of nearby objects. A long history of research shows that bees discriminate visually in ways that depend on their activity or task, so we must distinguish between vision during flying, fixating or hovering and landing. Bees return again and again for a reward of sugar solution and use their eyes to find their way. In an apparatus that makes them discriminate between two simulataneously visible but regularly interchanged targets, seen at a distance of 27 cm, bees are able to distinguish a remarkable number of simple patterns, but they fail in certain critical cases. The results can be explained with the hypothesis that bees have several broadly tuned overlapping filters with large fields that respond to the predominant orientation in a region of the image, and others for radial and circular patterns. Together with colour, these filters are independent of range. Bees prefer to use landmarks where they can, then global pattern at the largest scale, and lastly the detail around the goal. The way that discrimination of one visual feature is independent of other variables can be explained by models analogous to the colour triangle in colour discrimination.     

Visual discrimination by the honeybee (Apis mellifera): the position of the common centre as the cue

Abstract. Bees can be trained to discriminate between a target with a 20° spot above a 10° spot of the same colour, and another target with the spots exchanged in position. Tests show that they do not remember the separate positions of spots of the same colour (including black) on the same target. The bees discriminate the difference in positions, in the vertical direction, of the common centres of the spots taken together, with or without green contrast.
Similar results are obtained in discriminations of a fixed T shape, each composed of two broad black bars subtending 8 by 24°, vs the same shape inverted. The trained bees fail to discriminate between the T shapes when the centroids are at the same level in the vertical direction. Moving the shapes in the horizontal direction in tests has less effect. Quite different results are obtained when the two bars of the T shape differ in colour. The bees discriminate the positions of the two colours separately, but they still fail to discriminate the shape of the T. The results can be explained by filters that detect the intensities within their fields, irrespective of shape, and weigh them according to their vertical angles from the horizontal midline. The normal function of these filters could be to detect the levels of objects relative to the horizon when the bee is in flight.     

The forest or the trees: preference for global over local image processing is reversed by prior experience in honeybees

Traditional models of insect vision have assumed that insects are only capable of low-level analysis of local cues and are incapable of global, holistic perception. However, recent studies on honeybee (Apis mellifera) vision have refuted this view by showing that this insect also processes complex visual information by using spatial configurations or relational rules. In the light of these findings, we asked whether bees prioritize global configurations or local cues by setting these two levels of image analysis in competition. We trained individual free-flying honeybees to discriminate hierarchical visual stimuli within a Y-maze and tested bees with novel stimuli in which local and/or global cues were manipulated. We demonstrate that even when local information is accessible, bees prefer global information, thus relying mainly on the object''s spatial configuration rather than on elemental, local information. This preference can be reversed if bees are pre-trained to discriminate isolated local cues. In this case, bees prefer the hierarchical stimuli with the local elements previously primed even if they build an incorrect global configuration. Pre-training with local cues induces a generic attentional bias towards any local elements as local information is prioritized in the test, even if the local cues used in the test are different from the pre-trained ones. Our results thus underline the plasticity of visual processing in insects and provide new insights for the comparative analysis of visual recognition in humans and animals.     

Visual discriminations of spokes, sectors, and circles by the honeybee (Apis mellifera)

A new cue for visual discrimination by the honeybee has been demonstrated. Bees detected the position of the centre of symmetry of radial patterns of spokes, sectors, and circles relative to their point of choice in the learning process, irrespective of the pattern. When trained with one of these patterns versus a blank target, the bees discriminated a shift in the position of the centre of symmetry by as little as 5 degrees , in some cases with unfamiliar test patterns. A pattern of spokes or rings also stabilized the vision of the bees in the horizontal plane so that the position of a plain black area could then be discriminated. In other experiments, bees discriminated half of a pattern of radial spokes or concentric circles from the other half, cut either vertically or horizontally, and irrespective of scale. Therefore these patterns were not detected by preformed combinations of orientation detectors or global templates with a single output. Instead, the crucial cue for detecting edges as radial or circular was the coincidence of responses of numerous local edge detectors having the appropriate convergence to a hub. Edges that converged towards a hub were detected by the bees as radial, and edges at right angles to these were parts of circles, irrespective of the actual pattern. Breaking the patterns of spokes or circles into rows of squares spoiled the discrimination if the squares were separately resolved. Alternatively, breaking the pattern into short bars that were separately resolved spoiled the discrimination when the bars subtended less than 3 degrees . The local feature detectors for spokes and circles therefore resembled those of the orientation detectors in being short, independent, and unable to span gaps of more than 3 degrees . In conclusion, radial and circular patterns were identified by the regional coincidences and convergence of local detectors of edge orientation, and the positions of the centres of symmetry were remembered as landmarks that helped locate the reward, but the patterns themselves were not remembered.     

The visual system of the honeybee (Apis mellifera): the maximum length of the orientation detector

Bees were trained to discriminate between two or more black bars and similar bars at right angles, presented on a vertical surface. The positions of the bars were shifted every 5 min to prevent their locations being used as cues. The experiments exploit the fact that bees do not discriminate the global orientation of a straight line of small black squares that are individually resolved, because the local responses to equal lengths of edges at right angles cancel out, and each square has no residual orientation cue. The experiments measure the resolution of this effect by control of the width of the gaps between the squares. At the limit the unit orientation detectors cannot span the gaps. Training with vertical or with horizontal bars in separate experiments, and testing with vertical or horizontal lines of squares, shows that the vertical gaps in horizontal rows are detected with better resolution than horizontal gaps in vertical rows. The results show that unit orientation detectors span not more than 3 ommatidia.     

Two-dimensional pattern discrimination by the honeybee

For a reward of sugar, bees will learn to prefer a pattern rather than an alternative similar one. This visual discrimination allows us to measure resolution, and to search for the cues that the bees remember and later use to recognize the rewarded pattern. Two systems in parallel, analogous to low pass and high pass filters, are distinguished. The first system discriminates the location and size of at least one area of contrast on each side of the target, with inputs from blue and green receptors, but the ability to discriminate the location of colour depends upon fixation. The bees remember less than a low resolution copy of the image, even when they fixate on a vertical pattern. The second system amplifies the contrast at edges in the pattern, ignoring the direction of contrast, and controls fixation upon the target. Edges are discriminated according to their orientation and radial or tangential arrangement. An axis of bilateral symmetry is detected. However, the relative locations of cues within the image are lost, apparently because the relevant neurones have very large fields. Only the cues, not the whole patterns, are preserved in memory. This system is colour blind because its input is restricted to the receptors with peak sensitivity in the green. The two systems together discriminate many simple patterns, but not all, because the filters are limited in variety.     

Pattern discrimination by the honeybee (Apis mellifera): training on two pairs of patterns alternately

Pattern discrimination in the honeybee was studied by training alternately with two different pairs of patterns. Individually marked bees made a forced choice from a fixed distance in a standard Y-choice maze for a reward of sugar solution. Bees were trained, first on one pair of patterns for 10min then on a second pair, and so on, alternately between the two pairs. The pairs of patterns were selected to test the hypothesis that bees have a limited number of parallel mechanisms for the detection and discrimination of certain generalized global features. If this is so, it might be expected that each channel could process one pair of patterns simultaneously, but two pairs of patterns that are processed by the same channel would interfere with each other during the learning process. Features tested were: average orientation of edges, radial and tangential edges based on a symmetry of three or six, the position of a black spot, and the exchange of black and white. The bees fail to learn when the two alternated pairs of patterns offer the same feature, and they discriminate when the pairs offer two different features.