Search

Abstract Premise Automated disease, weed, and crop classification with computer vision will be invaluable in the future of agriculture. However, existing model architectures like ResNet, EfficientNet, and ConvNeXt often underperform on smaller, specialised datasets typical of such projects. Methods We address this gap with informed data collection and the development of a new convolutional neural network architecture, PhytNet. Utilising a novel dataset of infrared cocoa tree images, we demonstrate PhytNet's development and compare its performance with existing architectures. Data collection was informed by spectroscopy data, which provided useful insights into the spectral characteristics of cocoa trees. Cocoa was chosen as a focal species due to the diverse pathology of its diseases, which pose significant challenges for detection. Results ResNet18 showed some signs of overfitting, while EfficientNet variants showed distinct signs of overfitting. By contrast, PhytNet displayed excellent attention to relevant features, almost no overfitting, and an exceptionally low computation cost of 1.19 GFLOPS. Conclusions We show that PhytNet is a promising candidate for rapid disease or plant classification and for precise localisation of disease symptoms for autonomous systems. We also show that the most informative light spectra for detecting cocoa disease are outside the visible spectrum and that efforts to detect disease in cocoa should be focused on local symptoms, rather than the systemic effects of disease.

Abstract Plant pathogens can decimate crops and render the local cultivation of a species unprofitable. In extreme cases this has caused famine and economic collapse. Timing is vital in treating crop diseases, and the use of computer vision for precise disease detection and timing of pesticide application is gaining popularity. Computer vision can reduce labour costs, prevent misdiagnosis of disease, and prevent misapplication of pesticides. Pesticide misapplication is both financially costly and can exacerbate pesticide resistance and pollution. Here, we review the application and development of computer vision and machine learning methods for the detection of plant disease. This review goes beyond the scope of previous works to discuss important technical concepts and considerations when applying computer vision to plant pathology. We present new case studies on adapting standard computer vision methods and review techniques for acquiring training data, the use of diagnostic tools from biology, and the inspection of informative features. In addition to an in‐depth discussion of convolutional neural networks (CNNs) and transformers, we also highlight the strengths of methods such as support vector machines and evolved neural networks. We discuss the benefits of carefully curating training data and consider situations where less computationally expensive techniques are advantageous. This includes a comparison of popular model architectures and a guide to their implementation.

Abstract The extended female postreproductive life span found in humans and some toothed whales remains an evolutionary puzzle. Theory predicts demographic patterns resulting in increased female relatedness with age (kinship dynamics) can select for a prolonged postreproductive life span due to the combined costs of intergenerational reproductive conflict and benefits of late‐life helping. Here, we test this prediction using >40 years of longitudinal demographic data from the sympatric yet genetically distinct killer whale ecotypes: resident and Bigg's killer whales. The female relatedness with age is predicted to increase in both ecotypes, but with a less steep increase in Bigg's due to their different social structure. Here, we show that there is a significant postreproductive life span in both ecotypes with >30% of adult female years being lived as postreproductive, supporting the general prediction that an increase in local relatedness with age predisposes the evolution of a postreproductive life span. Differences in the magnitude of kinship dynamics however did not influence the timing or duration of the postreproductive life span with females in both ecotypes terminating reproduction before their mid‐40s followed by an expected postreproductive period of about 20 years. Our results highlight the important role of kinship dynamics in the evolution of a long postreproductive life span in long‐lived mammals, while further implying that the timing of menopause may be a robust trait that is persistent despite substantial variation in demographic patterns among populations.

Group navigation is of great importance for many animals, such as migrating flocks of birds or shoals of fish. One theory states that group membership can improve navigational accuracy compared to limited or less accurate individual navigational ability in groups without leaders (“Many-wrongs principle”). Here, we simulate leaderless group navigation that includes social connections as preferential interactions between individuals. Our results suggest that underlying social networks can reduce navigational errors of groups and increase group cohesion. We use network summary statistics, in particular network motifs, to study which characteristics of networks lead to these improvements. It is networks in which preferences between individuals are not clustered, but spread evenly across the group that are advantageous in group navigation by effectively enhancing long-distance information exchange within groups. We suggest that our work predicts a base-line for the type of social structure we might expect to find in group-living animals that navigate without leaders [Current Zoology 58 (2): 329-341, 2012].

Prey that are unprofitable to attack (for example, those containing noxious chemicals) are often conspicuously patterned and move in a slower and more predictable manner than species lacking these defences. Contemporary theories suggest these traits have evolved as warning signals because they can facilitate both associative and discriminative avoidance learning in predators. However, it is unclear why these particular traits and not others have tended to evolve in unprofitable prey. Here we show using a signal detection model that unprofitable prey will evolve conspicuous colours and patterns partly because these characteristics cannot readily evolve in profitable prey without close mimicry. The stability of this signal is maintained through the costs of dishonesty in profitable prey. Indeed, unprofitable prey will sometimes evolve a conspicuous form to reduce mimetic parasitism, even in the unlikely event that this form can be more closely mimicked. This is one of the first mathematical models of the evolution of warning signals to allow for the possibility of mimicry, yet our analyses suggest it may offer a general explanation as to why warning signals take the form that they do. Warning signals and mimicry may therefore be more closely related than is currently supposed. [ABSTRACT FROM AUTHOR]

Mathematical models of mimicry typically involve artificial prey species with fixed colorations or appearances; this enables a comparison of predation rates to demonstrate the level of protection a mimic might be afforded. Fruitful theoretical results have been produced using this method, but it is also useful to examine the possible evolutionary consequences of mimicry. To that end, we present individual-based evolutionary simulation models where prey colorations are free to evolve. We use the models to examine the effect of Batesian mimics on Müllerian mimics and mimicry rings. Results show that Batesian mimics can potentially incite Müllerian mimicry relationships and encourage mimicry ring convergence. [ABSTRACT FROM AUTHOR]