Visual Signals Research Articles

DFCNet +: Cross-modal dynamic feature contrast net for continuous sign language recognition

In sign language communication, the combination of hand signs and facial expressions is used to convey messages in a fluid manner. Accurate interpretation relies heavily on understanding the context of these signs. Current methods, however, often focus on static images, missing the continuous flow and the story that unfolds through successive movements in sign language. To address this constraint, our research introduces the Dynamic Feature Contrast Net Plus (DFCNet+), a novel model that incorporates both dynamic feature extraction and cross-modal learning. The dynamic feature extraction module of DFCNet+ uses dynamic trajectory capture to monitor and record motion across frames and apply key features as an enhancement tool that highlights pixels that are critical for recognizing important sign language movements, allowing the model to follow the temporal variation of the signs. In the cross-modal learning module, we depart from the conventional approach of aligning video frames with textual descriptions. Instead, we adopt a gloss-level alignment, which provides a more detailed match between the visual signals and their corresponding text glosses, capturing the intricate relationship between what is seen and the associated text. The enhanced proficiency of DFCNet+ in discerning inter-frame details translates to heightened precision on benchmarks such as PHOENIX14, PHOENIX14-T and CSL-Daily. Such performance underscores its advantage in dynamic feature capture and inter-modal learning compared to conventional approaches to sign language interpretation. Our code is available at https://github.com/fyzjut/DFCNet_Plus.

Measuring and Predicting Multisensory Reaction Latency: A Probabilistic Model for Visual-Auditory Integration.

Virtual/augmented reality (VR/AR) devices offer both immersive imagery and sound. With those wide-field cues, we can simultaneously acquire and process visual and auditory signals to quickly identify objects, make decisions, and take action. While vision often takes precedence in perception, our visual sensitivity degrades in the periphery. In contrast, auditory sensitivity can exhibit an opposite trend due to the elevated interaural time difference. What occurs when these senses are simultaneously integrated, as is common in VR applications such as 360° video watching and immersive gaming? We present a computational and probabilistic model to predict VR users' reaction latency to visual-auditory multisensory targets. To this aim, we first conducted a psychophysical experiment in VR to measure the reaction latency by tracking the onset of eye movements. Experiments with numerical metrics and user studies with naturalistic scenarios showcase the model's accuracy and generalizability. Lastly, we discuss the potential applications, such as measuring the sufficiency of target appearance duration in immersive video playback, and suggesting the optimal spatial layouts for AR interface design.

Teaching labs for blind students: Equipment to measure the inertia of simple objects

This article describes a laboratory experiment for blind students to measure the moment of inertia of simple objects; in this case, that of a disc about its axis of symmetry. We adapted our usual lab to modify the data collection process, using an open-source electronic platform to convert visual signals into acoustic signals. This allowed blind students at our University to participate alongside their classmates in the mechanics lab.

Experience with aposematic defense triggers attack bias in a mantid predator (Stagmomantis carolina)

Abstract Studies of predator psychology in aposematism have suggested important effects of signal detection through space and time on outcomes of attack behavior. Both the integration of aposematic signals from prey and experience state of the predator can have important effects on attack decisions. The universality of these effects however, especially as it applies to non-avian predators such as arthropods, remains poorly understood. We examined the effects of multimodal aposematic signaling and prior experience with aposematism on attack latency and attack likelihood of the Carolina mantis (Stagmomantis carolina). Using artificial prey bearing visual and olfactory signals of the convergent lady beetle (Hippodamia convergens), we tested two cohorts of mantids (representing juvenile and adult stages) across four groups: visual only, odor only, combined signals, and control. We then used approaches in linear modeling to test the hypotheses that 1) prior experience with aposematism alters attack behavior toward aposematic prey and 2) multimodal signals have synergistic effects on attack behavior relative to either unisensory signal presented alone. We found support for the first hypothesis in that mantids employ attack biases against visual and olfactory aposematic signals, but only after prior exposure to aposematism and only as juveniles. While support is lacking for multimodal integration by the mantids, this study is the first to suggest a response of mantids to an aposematic olfactory signal (in addition to visual signal) and may suggest a developmental window for mantid predators to develop biases toward aposematic prey that are shaped by experience.

New Solution to 3D Projection in Human-like Binocular Vision

A human eye has about 120 million rod cells and 6 million cone cells. This huge number of light-sensing cells inside a human eye will continuously produce a huge quantity of visual signals that flow into the human brain for daily processing. However, the real-time processing of these visual signals does not cause excessive energy consumption by the human brain. This fact tells us the truth which is to say that human-like vision processes do not rely on complicated and expensive formulas to compute depth, displacement, and colors. On the other hand, the human eye is like a camera with pan-tilt (PT) motions. We all know that in computer vision, each set of PT parameters (i.e., coefficients of pan motion as well as tilt motion) requires a dedicated calibration to determine a camera’s projection matrix. Since there is an infinite number of PT parameters that could be produced by a human eye, it is unlikely that a human brain stores an infinite number of calibration matrices for each human eye. These observations inspire us to look for a simpler and computationally non-expensive solution which is to undertake three-dimensional (3D) projection in human-like binocular vision. In other words, it is an interesting question for us to answer, which is to say whether simpler and learning-friendly formulas for computing depth and displacement exist or not. If the answer is yes, these formulas must also be calibration-friendly (i.e., easy process on the fly or on the go). In this paper, we present an important discovery of a new solution to 3D projection in a human-like binocular vision system. This solution is computationally simpler and could be easily learned or calibrated on the fly. We know that the purpose of doing 3D projection in binocular vision is to undertake forward and inverse transformations (or mappings) between coordinates in 2D digital images and coordinates in a 3D analogue scene. The formulas underlying the new solution are accurate, easily computable, easily tunable (i.e., to be calibrated on the fly or on the go) and could be easily implemented by a neural system (i.e., a network of neurons or a network of computational flows). Experimental results have validated the newly derived formulas which are better than textbook solutions.

Interactions between sexual signaling and wing size drive ecology and evolution of wing colors in Odonata

Insect coloration has evolved in response to multiple pressures, and in Odonata (dragonflies and damselflies) a body of work supports a role of wing color in a variety of visual signals and potentially in thermoregulation. Previous efforts have focused primarily on melanistic coloration even though wings are often multicolored, and there has yet to be comprehensive comparative analyses of wing color across broad geographic regions and phylogenetic groups. Percher vs. flier flight-style, a trait with thermoregulatory and signaling consequences, has not yet been studied with regard to color. We used a new color clustering approach to quantify color across a dataset of over 8,000 odonate wing images representing 343 Nearctic species. We then utilized phylogenetically informed Bayesian zero-inflated mixture models to test how color varies with mean ambient temperature, body size, sex and flight-style. We found that wing coloration clustered into two groups across all specimens - light brown-yellow and black-dark brown - with black-dark brown being a much more cohesive grouping. Male perchers have a greater proportion of black-dark brown color on their wings as do species with longer wings. In colder climates, odonates were more likely to have black-dark brown color present, but we found no relationship between the proportion of black and temperature. Light brown-yellow showed similar scaling with wing length, but no relationship with temperature. Our results suggest that black-dark brown coloration may have a limited role in thermoregulation, while light brown-yellow does not have such a role. We also find that the odonate sexes are divergent in wing color in percher species only, suggesting a strong role for color in signaling in more territorial males. Our research contributes to an understanding of complex interactions driving ecological and evolutionary dynamics of color in animals.

Modeling saccade reaction time in marmosets: the contribution of earlier visual response and variable inhibition

Marmosets are expected to serve as a valuable model for studying the primate visuomotor system due to their similar oculomotor behaviors to humans and macaques. Despite these similarities, differences exist; challenges in training marmosets on tasks requiring suppression of unwanted saccades, having consistently shorter, yet more variable saccade reaction times (SRT) compared to humans and macaques. This study investigates whether the short and variable SRT in marmosets is related to differences in visual signal transduction and variability in inhibitory control. We refined a computational SRT model, adjusting parameters to better capture the marmoset SRT distribution in a gap saccade task. Our findings indicate that visual information processing is faster in marmosets, and that saccadic inhibition is more variable compared to other species.

Temporal dynamics of scent mark composition in field‐experimental lizard populations

Abstract Animal signals are dynamic traits that can undergo considerable spatial and temporal changes and that are influenced by factors such as age, health condition and interactions with both the abiotic and biotic environment. However, much of our understanding of signal changes throughout an individual's lifetime stems from cross‐sectional, often laboratory‐based, studies focused on visual and auditory signals. Longitudinal field investigations of temporal variation in chemical signals, especially in vertebrates, remain rare despite chemical communication being the most ubiquitous form of information exchange in the natural world. To remedy this, we conducted a unique, replicated field experiment to study the temporal signal dynamics in free‐living lizard populations on natural islands. Specifically, we collected scent marks from individually marked lizards across five populations during the spring of two consecutive years and analysed the lipophilic chemical composition of these scent marks. Our findings demonstrate that the overall scent mark composition of individual lizards changed over time, shifting consistently in both direction and magnitude from year to year among individuals and across replicate populations. Similar patterns were observed for the chemical richness and diversity of scent marks. Temporal variation in the relative proportions of three potentially socially relevant signalling compounds in lizard scent marks revealed a more complex pattern: α‐tocopherol remained stable over time, oleic acid decreased and the change in octadecanoic acid proportion was body size‐dependent. Together, our results provide novel insights into how individual vertebrate chemical signals may fluctuate across space and time. We discuss the potential causes of the observed temporal variability and its consequences for chemical signal evolution. Read the free Plain Language Summary for this article on the Journal blog.

Visual signal transduction for environmental stewardship: A novel biosensing approach to identify and quantify chlorpyrifos-related residues in aquatic environments

The pervasive presence of organophosphate pesticides (OPs), such as chlorpyrifos (CPF), in aquatic ecosystems underscores the urgent need for sensitive and reliable detection methods to safeguard environmental and public health. This study addressed the critical need for a novel biosensor capable of detecting CPF and its toxic metabolite, 3,5,6-trichloro-2-pyridinol (TCP), with high sensitivity and selectivity, suitable for field applications in environmental monitoring. The study engineered a whole-cell biosensor based on E. coli strains that utilize the ChpR transcriptional regulator and the vioABCE gene cluster, providing a distinct visual and colorimetric response to CPF and TCP. The biosensor's performance was optimized and evaluated across various water matrices, including freshwater, seawater, and soil leachate. The biosensor demonstrated high sensitivity with a broad linear detection range, achieving limits of detection (LODs) at 0.8 μM for CPF and 7.813 nM for TCP. The linear regression concentration ranges were 1.6–12.5 μM for CPF and 15.6–125 nM for TCP, aligning with environmental standard limits and ensuring the biosensor's effectiveness in real-world scenarios. This innovative biosensing approach offers a robust, user-friendly tool for on-site environmental monitoring, significantly mitigating OPs contamination and advancing current detection technologies to meet environmental protection standards.

Inspiration from Visual Ecology for Advancing Multifunctional Robotic Vision Systems: Bio-inspired Electronic Eyes and Neuromorphic Image Sensors.

In robotics, particularly for autonomous navigation and human-robot collaboration, the significance of unconventional imaging techniques and efficient data processing capabilities is paramount. The unstructured environments encountered by robots, coupled with complex missions assigned to them, present numerous challenges necessitating diverse visual functionalities, and consequently, the development of multifunctional robotic vision systems has become indispensable. Meanwhile, rich diversity inherent in animal vision systems, honed over evolutionary epochs to meet their survival demands across varied habitats, serves as a profound source of inspirations. Here, recent advancements in multifunctional robotic vision systems drawing inspiration from natural ocular structures and their visual perception mechanisms are delineated. First, unique imaging functionalities of natural eyes across terrestrial, aerial, and aquatic habitats and visual signal processing mechanism of humans are explored. Then, designs and functionalities of bio-inspired electronic eyes are explored, engineered to mimic key components and underlying optical principles of natural eyes. Furthermore, neuromorphic image sensors are discussed, emulating functional properties of synapses, neurons, and retinas and thereby enhancing accuracy and efficiency of robotic vision tasks. Next, integration examples of electronic eyes with mobile robotic/biological systems are introduced. Finally, a forward-looking outlook on the development of bio-inspired electronic eyes and neuromorphic image sensors is provided.

Investigating premotor corticospinal excitability in fast and slow voluntary contractions of the elbow flexors.

Corticospinal excitability (CSE) increases prior to a voluntary contraction; however, the relative contributions of premotor cortical and spinal mechanisms are poorly understood. It is unknown whether the intended voluntary contractile rate affects CSE. Eighteen young, healthy participants (nine females) completed isometric elbow flexion contractions targeting 50% maximal voluntary contraction (MVC) torque, at either fast (fast as possible) or slow (25% MVC/s) contractile rates. Participants were cued to contract with warning (red) and "GO" (green) visual signals. Magnetic and electric stimulations were applied to elicit motor evoked potentials (MEPs), cervicomedullary evoked potentials (CMEPs), and M-waves, in the surface electromyogram (EMG) recorded over the biceps brachii. MEPs and CMEPs were collected at 0, 25, 50 and 75% premotor reaction time (RT - defined as the time between the "GO" cue and onset of biceps brachii EMG) and compared to a resting baseline. MEP amplitude was greater than baseline at 75% RT (p=0.009), and CMEP amplitude was significantly increased at all RT points relative to baseline (p≤0.001). However, there were no differences in MEP and CMEP amplitudes when compared between fast and slow conditions (p≥0.097). Normalized to the CMEP, there was no difference in MEP amplitude from baseline in either contractile condition (p≥0.264). These results indicate that increased premotor CSE is a spinally-mediated response. Furthermore, premotor CSE is not influenced by the intended voluntary contractile rate. CMEP amplitudes were larger for females than males within the premotor RT period (p=0.038), demonstrating that premotor spinal excitability responses may be influenced by sex.

A highly sensitive Photothermal Immunochromatographic test strip for detection of aflatoxin B1 based on CoS@TA Nanospheres

A novel photothermal immunochromatographic test strip (PITS) with tannic acid (TA) modified cobalt sulfide (CoS) nanospheres (CoS@TA) as immuno-probe element was developed for the detection of aflatoxin B1 (AFB1). CoS nanospheres (CoS NPs) with excellent photothermal conversion efficiency (η = 47.5 %) was synthesized and skillfully chelated with TA to improve its dispersion, biocompatibility, and chromatographic properties. After modification, the CoS@TA coupled with monoclonal antibody (mAb) against AFB1 (CoS@TA-mAb) by simple physical adsorption. The CoS@TA based PITS achieved highly sensitive detection of AFB1 with the limit of detection in photothermal signal (photothermal-LOD) of 0.00503 μg/L, which was 19.88-fold higher than the LOD in visual signal (visual-LOD, 0.1 μg/L). The application of TA in the modification of CoS provided ideas to improve the properties of functional nanomaterials such as dispersion and biocompatibility, and the application of CoS@TA in PITS construction laid a methodological foundation for further improving the detection sensitivity of trace targets.

Subjective and objective quality evaluation for industrial images

Recently, the demand for ever-better image processing technique continues to grow in field of industrial scenario monitoring and industrial process inspection. The subjective and objective quality evaluation of industrial images are vital for advancing the development of industrial visual perception and enhancing the quality of industrial image/video processing applications. However, the scarcity of publicly available industrial image databases with reliable subjective scores restricts the development of industrial image quality evaluation (IIQE). In preparation for a vacancy, this article first establishes two industrial image databases (i.e., industrial scenario image dataset (ISID) and industrial process image dataset (IPID)) for assessing IIQE metrics. Furthermore, in order to avoid overwhelming industrial image nuances due to the wavelet subband summation, we then present a novel industrial application subband information fidelity standard (SIFS) evaluation method using the channel capacity of visual signals in wavelet domain. Specifically, we first build a visual signals channel model based on perception process from human eyes to brain. Second, we compute and compare the channel capacity for reference and distorted images to measure the information fidelity in each wavelet subband. Third, we sum over the subbands for information fidelity ratio to obtain the overall quality score. Finally, we fairly compare some up-to-date and our proposed image quality evaluation (IQE) methods in two novelty industrial datasets respectively. Our ISID and IPID datasets are capable of evaluating most IQE metrics comprehensively and paves the way for further research on IIQE. Our SIFS model show a remarkable performance comparing with other up-to-date IQE methods.

Nanozyme-Based Colorimetric and Smartphone Imaging Advanced Sensing Platforms for Tetracycline Detection and Removal in Food

The presence of antibiotic residues poses a significant threat to food assurance, triggering widespread concerns. Therefore, the prompt and accurate detection and removal of antibiotic residues are essential for ensuring food safety. In this study, an aptmer modified triple-metal nanozyme (apt-TMNzyme) sensor was developed, which achieved a portable, visual, intelligent, and fast determination for tetracycline (TET). The proposed apt-TMNzyme exhibited willow leaf-like morphology, high specific surface area and excellent TET adsorption and removal properties. The experiments showed that the apt-TMNzyme had outstanding peroxidase activity and could catalyze the oxidation of 3,3',5,5'-tetramethylbenzidine (TMB) to produce a blue product in the presence of H2O2, which provided a visual response signal to TET. This sensor was capable of quantifying TET within a concentration range of 0.2 nM-70 μM, achieving a detection limit of 7.1 nM under optimal conditions. When tested on real food samples, our sensor produced results that closely paralleled those achieved through high-performance liquid chromatography. To improve accessibility and user-friendliness, we also designed a colorimetric testing paper integrated with a smartphone application for intuitive and intelligent detection of TET, which enables the quantitative determination of TET in the concentration range of 0.003-60 μM, the detection limit was 5.1 μM. This integrated portable sensor not only streamlines the testing process, saving time and costs, but also offers a promising solution for rapid and sensitive detection of antibiotic residues.

CalTrig: A GUI-Based Machine Learning Approach for Decoding Neuronal Calcium Transients in Freely Moving Rodents.

Advances in in vivo Ca2+ imaging using miniatured microscopes have enabled researchers to study single-neuron activity in freely moving animals. Tools such as MiniAN and CalmAn have been developed to convert Ca2+ visual signals to numerical information, collectively referred to as CalV2N. However, substantial challenges remain in analyzing the large datasets generated by CalV2N, particularly in integrating data streams, evaluating CalV2N output quality, and reliably and efficiently identifying Ca2+ transients. In this study, we introduce CalTrig, an open-source graphical user interface (GUI) tool designed to address these challenges at the post-CalV2N stage of data processing. CalTrig integrates multiple data streams, including Ca2+ imaging, neuronal footprints, Ca2+ traces, and behavioral tracking, and offers capabilities for evaluating the quality of CalV2N outputs. It enables synchronized visualization and efficient Ca2+ transient identification. We evaluated four machine learning models (i.e., GRU, LSTM, Transformer, and Local Transformer) for Ca2+ transient detection. Our results indicate that the GRU model offers the highest predictability and computational efficiency, achieving stable performance across training sessions, different animals and even among different brain regions. The integration of manual, parameter-based, and machine learning-based detection methods in CalTrig provides flexibility and accuracy for various research applications. The user-friendly interface and low computing demands of CalTrig make it accessible to neuroscientists without programming expertise. We further conclude that CalTrig enables deeper exploration of brain function, supports hypothesis generation about neuronal mechanisms, and opens new avenues for understanding neurological disorders and developing treatments.

Substance P and dopamine form a "push-pull" system that diurnally regulates retinal gain.

The operation of the retina, like other brain circuits, is under modulatory control. One coordinator of changes in retinal function is dopamine, a neuromodulator released in a light-dependent way to adjust vision on a diurnal cycle. Here, we demonstrate that substance P is a similarly powerful retinal modulator that interacts with the dopamine system. By imaging glutamatergic synaptic transmission in larval zebrafish, we find that substance P decreases the contrast sensitivity of ON and OFF visual channels up to 8-fold, with suppression of visual signals being strongest through the "transient" pathway responding to higher frequencies. These actions are exerted in the morning, in large part by suppressing the amplification of visual signals by dopamine, but substance P is almost completely inactive in the afternoon. Modulation of retinal gain is accompanied by changes in patterns of vesicle release at the synapses of bipolar cells: increased gain shifts coding of stimulus strength from the rate of release events to their amplitude generated by a process of multivesicular release (MVR). Together, these actions of substance P reduce the flow of visual information, measured in bits, ∼3-fold. Thus, whereas dopamine "pushes" the retina to transmit information at higher rates in the afternoon, substance P acts in antiphase to suppress dopamine signaling and "pull down" information transmission in the morning.

OODNet: A deep blind JPEG image compression deblocking network using out-of-distribution detection

JPEG is one of the most popular image compression techniques, with numerous applications ranging from medical imaging to surveillance systems. Since JPEG introduces the blocking artifacts to the decompressed visual signals, enhancing the quality of these images is of paramount importance. Recently, various deep neural networks have been proposed for JPEG image deblocking that can effectively reduce the blocking artifacts produced by the JPEG compression technique. However, most of these schemes could only handle decompressed images generated by a set of specific JPEG quality factor (QF) values employed in the network training process. Therefore, when the images are obtained by the JPEG QF values other than those used in the network training process, the performance of deep learning-based JPEG image deblocking schemes drops significantly. To address this, in this paper, we propose a novel deep learning-based blind JPEG image deblocking method, which employs out-of-distribution detection to perform deblocking efficiently for various quality factor (QF) values. The proposed scheme can distinguish between the decompressed images using the QF values used in the training set and those using the QF values not used in the training set, and then, a suitable deblocking strategy for generating high-quality images is developed. The proposed scheme is shown to outperform the state-of-the-art JPEG image deblocking methods for various QF values.

Concentration Techniques in the Classroom for Children with ADHD, in an Educational Unit in the Province of Santa Elena

This study analyzes the effectiveness of various concentration techniques applied in classrooms for children with Attention Deficit Hyperactivity Disorder (ADHD) in an educational unit in Santa Elena. Employing a mixed-methods approach, surveys and interviews with teachers were examined, and the behavior of students with ADHD was observed. The findings suggest that strategies like environmental adaptation, use of visual and auditory signals, and positive reinforcement are effective in enhancing concentration and academic performance of these students. It is concluded that the implementation of specific concentration techniques, tailored to the size and resources of each institution, is crucial to support students with ADHD, thereby improving their academic performance and satisfaction.

Exploring the categorical nature of colour perception: Insights from artificial networks

The electromagnetic spectrum of light from a rainbow is a continuous signal, yet we perceive it vividly in several distinct colour categories. The origins and underlying mechanisms of this phenomenon remain partly unexplained. We investigate categorical colour perception in artificial neural networks (ANNs) using the odd-one-out paradigm. In the first experiment, we compared unimodal vision networks (e.g., ImageNet object recognition) to multimodal vision-language models (e.g., CLIP text-image matching). Our results show that vision networks predict a significant portion of human data (approximately 80%), while vision-language models account for the remaining unexplained data, even in non-linguistic experiments. These findings suggest that categorical colour perception is a language-independent representation, though it is partly shaped by linguistic colour terms during its development. In the second experiment, we explored how the visual task influences the colour categories of an ANN by examining twenty-four Taskonomy networks. Our results indicate that human-like colour categories are task-dependent, predominantly emerging in semantic and 3D tasks, with a notable absence in low-level tasks. To explain this difference, we analysed kernel responses before the winner-takes-all stage, observing that networks with mismatching colour categories may still align in underlying continuous representations. Our findings quantify the dual influence of visual signals and linguistic factors in categorical colour perception and demonstrate the task-dependent nature of this phenomenon, suggesting that categorical colour perception emerges to facilitate certain visual tasks.

Channel randomisation: Self-supervised representation learning for reliable visual anomaly detection in speciality crops

Modern, automated quality control systems for speciality crops utilise computer vision together with a machine learning paradigm exploiting large datasets for learning efficient crop assessment components. To model anomalous visuals, data augmentation methods are often developed as a simple yet powerful tool for manipulating readily available normal samples. State-of-the-art augmentation methods embed arbitrary “structural” peculiarities in normal images to build a classifier of these artefacts (i.e., pretext task), enabling self-supervised representation learning of visual signals for anomaly detection (i.e., downstream task). In this paper, however, we argue that learning such structure-sensitive representations may be suboptimal for agricultural anomalies (e.g., unhealthy crops) that could be better recognised by a different type of visual element like “colour”.To be specific, we propose Channel Randomisation (CH-Rand)—a novel data augmentation method that forces deep neural networks to learn effective encoding of “colour irregularities” under self-supervision whilst performing a pretext task to discriminate channel-randomised images. Extensive experiments are performed across various types of speciality crops (apples, strawberries, oranges, and bananas) to validate the informativeness of learnt representations in detecting anomalous instances. Our results demonstrate that CH-Rand’s representations are significantly more reliable and robust, outperforming state-of-the-art methods (e.g., CutPaste) that learn structural representations by over 43% in Area Under the Precision–Recall Curve (AUC–PR), particularly for strawberries. Additional experiments suggest that adopting the L∗a∗b∗ colour space and “curriculum” learning in the pretext task — gradually disregarding channel combinations for unrealistic outcomes — further improves downstream-task performance by 16% in AUC–PR. In particular, our experiments employ Riseholme-2021, a novel speciality crop dataset consisting of 3.5K real strawberry images gathered in situ from the real farm, along with the Fresh & Stale public dataset. All our code and datasets are made publicly available online to ensure reproducibility and encourage further research in agricultural technologies.