Global warming causes coral bleaching which threatens the health and existence of coral reefs and therefore also the future of a lot of species, including human beings. Efforts to automate coral reef monitoring using annotated coral images to detect coral bleaching are hindered by the lack of a complete dataset that specifies the health and bleaching status of corals. We propose to combine publicly available data into a dataset and train a CNN for coral bleaching detection. This model performs surprisingly well. However, combining data from different sources gives rise to dataset biases which helps classifiers perform better and make them unreliable for unseen data. We try to detect such biases and document themsing several bias detection methods.

The Routing Protocol for Low-Power and Lossy Networks (RPL) has gained in popularity since the increased connectivity of everyday items to the Internet. One of the discovered attacks on RPL is the rank attack, which opens up possibilities for attackers to control traffic in the RPL network by spoofing their priority. Many solutions have been proposed to mitigate this attack over the past few years. There is no perfect solution yet, partly because the success of a mitigation is dependent on the network configuration in which it is implemented. In some network configurations,as this paper will show, common mitigation solutions are less effective. By selecting and analyzing four well-cited mitigation and detection solutions, the effectiveness of these proposals is reviewed when the network is configured to use nonlinear objective functions (NOFs). After this, a proposal is given to defend against a rank attack when using NOFs. TRAIL was proposed as a solution for preventing decreased rank attacks and uses a challenge-response mechanism to verify the path from a node to the root. This paper proposes T-TRAIL; an extended version of TRAIL that allows the measurement of downwards-trip-time to detect outliers in the network. By doing this, the rank attack can be prevented when the network uses a NOF. Finally, an estimation of the performance impact of T-TRAIL on the network is given based on the performance measurements of TRAIL.

Routing Protocol for Low Power and Lossy Networks (RPL) is a routing protocol for Internet of Things (IoT) devices with limited resources. As IoT is becoming prevalent, it is important to secure the underlying protocols that compose it such as RPL. This paper sought to avoid an RPL-specific routing attack by modifying the protocol's functionalities. As a result, STIR is a novel method improving memory efficiency of routing tables in RPL's storing mode of operation to prevent Routing Table Overload attacks. STIR assigns IPv6 addresses in a specific way to form clusters of addresses, thus resulting in routing tables with sizes proportional to the node's number of sub-DODAGs. The proposed contribution increases memory efficiency in RPL's storing mode, therefore preventing routing table overload attacks. STIR necessitates few protocol modifications and additional control messages making it an ideal preventive method.

While the Routing Protocol for Low Power and Lossy Networks (RPL) is built to support the constraints of low-powered devices, it struggles to meet the standards in security. Generally, low-powered devices are challenged with limited cryptography, tough key management, and interoperability issues. Despite these concerns, security is not only deficient for RPL, but proposed mitigations appear untouched as well. This paper therefore contributes a lightweight cryptosystem. It questions and justifies its virtue in a twofold. First, we illustrate the importance of this mitigation with an impactful Sybil attack that enables malicious routing on root level. Second, we construct an attack pattern model and life-cycle to demonstrate the operational capabilities and objectives of the adversary for Internet of Things (IoT) generic attack patterns. The cryptosystem divides IEEE 802.15.4 networks into isolated clusters with key derivation functions. Because the key derivation function adopts spectrum resource measurements, the proposal includes a cooperative defence to validate these measurements of joining nodes. To avoid overhead, the mitigation operates on symmetric-key cryptography, piggybacked cluster identifiers, and Maximum Transmission Unit (MTU) requests to the trusted party that stores encrypted identity keys of member nodes. This mitigation, when combined with the efficient routing of RPL, enables a broad application for smart low-power constrained devices in a scalable IoT network while it protects against Sybil attacks and eavesdropping.

The IPv6 routing protocol for low-power and lossy networks (RPL) is a routing protocol that is standardized for constrained devices. This standard only considers static nodes and consequently underperforms in networks with moving nodes. Several studies exist intending to mend this problem, but analyses of RPL's performance in mobile situations are too scarce. These studies are needed to help researchers find related future work directions and improve RPL's support for moving nodes. This paper, therefore, analyzes the performance of RPL in dynamic networks and compares this to its performance in static networks by considering several routing and security metrics. It focuses on the impact of mobility when a node joins, leave, or moves within the network. The analysis concludes by discussing the effects of dynamic nodes on larger scale networks, DODAG's with multiple roots and networks with mobile roots. Afterward, DIS flooding is considered as an example of how attacks and their mitigations can be impacted by mobility, showing that more work is needed to secure RPL in these situations. This paper is constructed with a literature review and includes no experiments as the analyses in this research are broader than a few testable configurations.

Transformers can generate predictions auto-regressively by conditioning each sequence element on the previous ones, or can produce output sequences in parallel. While research has mostly explored upon this difference on tasks that are sequential in nature, we study this contrast on visual set prediction tasks, to analyze the core behaviour of the Transformer model. Multi-label classification, object detection and polygonal shape prediction are all visual set prediction tasks. Precisely predicting polygons in images is an important set prediction problem because polygons are representative of numerous types of objects, such as buildings, people, or obstacles for aerial vehicles. Set prediction is a difficult challenge for deep learning architectures as sets can have different cardinalities and are permutation invariant. We provide evidence on the importance of natural orders for Transformers, analyze the strengths and weaknesses of different solutions that can solve the set prediction task directly, and show the benefit of decomposing complex polygons into sets of ordered points in an auto-regressive manner.

Wheat is a widely used ingredient for food products. To increase the productionand quality of wheat, the density of ’wheat heads’ in a farm can be studied. Accuratelylocating wheat heads in images can be challenging. A lot of work has taken place insupervised semantic segmentation, but these networks typically require large pixel-wisehuman-annotated labeled data. Gathering this data is tedious and labour intensive.This paper proposes to use the novel unsupervised semantic segmentation model W-Net to solve this problem. To improve the accuracy, we investigated the influence ofthe frequency domain, by pre-processing the training data two different times using acustom filter, based on frequencies found in wheat heads, and a high pass filter.The approach is evaluated on the Global Wheat Head Detection (GWHD) dataset[11]. To compare the accuracy the generated segmentations were mapped to boundingboxes based. The proposed method did not show to be able to generate competing de-tection compared to the baseline method associated with the GWHD dataset, but theGWHD dataset has a different measurement of truth, consisting out bounding boxesinstead of segments which is in the disadvantage for the W-Net.Pre-processing the dataset using the high pass filter did increase the intersection overunion with 1,4% and the deviation of the reconstruction loss was smaller when fre-quency filtering was applied.Although the object detection has a low accuracy, this study showed that some ba-sic wheat head detection can be achieved by using the unsupervised segmentationmethod W-Net and the accuracy can be increased if a high pass filter is applied aspre-processing step.

Convolutional Neural Networks (CNNs) have made significant strides in the field of image processing over the last decade. Different approaches have been taken and improvements have been suggested. This paper looks at a newer novelty to neural networks for image counting, which is based on single-pixel center localization instead of the traditional bounding boxes. This neural network’s loss function is the weighted average Hausdorff distance, which does not only take into account the number of misclassified points but also the distance between predicted points and ground truth values. The paper aims to compare the accuracy of the single-pixel center neural network on original training images of wheat heads as compared to filtered images. The filtered images have had a band pass filter applied to them, that is constructed by looking at the average frequency of wheat heads. It filters out certain lower and higher frequencies up to a threshold, and its aim is to reduce background noise and accentuate the wheat heads. Results showed that there was no significant and attributable improvement in the performance of the object counter when trained on images with filtered frequency information. A discussion of the unexpected results then carries out, with the aim of rationalizing the insignificant improvement in performance of the neural network on filtered images. As part of the discussion and conclusion, a recommendation is also made, giving insights into determining if this single-pixel center neural network is appropriate for a given dataset of images.

How do automatic object detector outputs align with what humans consider good object detection? Our study is based on the responses of 70 participants for a survey. The participants are presented with images having bound- ing box predictions, their task is to choose images which according to them have an acceptable or a good detection. The results show a correlation between the size of the object and the evaluation metric IoU (Intersection over Union), with the size of the bounding box. Furthermore, the data indicates that the kind of box they prefer most for a detection output, is also the most accepted detection by them. Additionally, the results suggest that based on the symmetry of the object, position of the bounding box may or may not play a role for considering a detection valid. Our study investigates through human subjective choices if the traditional threshold value of IoU for evaluation, and tight bounding box outputs are always the best outputs in object detection techniques.

While sailing, sailors solely rely on their eyes to inspect the sail shape and adjust the sail configurations to achieve an appropriate sail shape that corresponds to the weather condition. This so-called trimming process requires years of experience. Hence the visual inspection of sail shape suffers from inaccuracy and communication difficulties. Therefore, this research proposes a visual analysis tool that presents an accurate sail shape structure and supports sailors in investigating the optimal sail shape of certain weather conditions. In order to achieve our goals, we reconstruct the sail shape using two triangulation methods. For incomplete reconstructed sail shapes caused by the uncertain image quality, we deform a complete template sail under the constraints of the incomplete sail shape and use the deformed sail as an estimation of the proper shape structure. We designed a visualization dashboard for sailors to explore the 3D structure, 2D profiles and characteristics of the time-varying sail shape and analyze their relation to boat speed. The usability of the visualization tool is tested by a qualitative evaluation with two sailing experts. The result shows that the reconstruction and deformation of sail shape are valid, and the visualization dashboard has the potential to enhance sailors' comprehension of sail shape and provide insights into the optimal sail shape trimming.

Person re-identification based on appearance is challenging due to varying views and lighting conditions in different cameras, or when multiple persons wear similar clothing styles and color. Considering these challenges, gait patterns provide an alternative to appearance, as gait can be captured from a distance and at a low resolution. In this paper we investigate and evaluate running gait as a unique attribute for video person re-identification in a recreational long-distance running event with 257 participants. We show that running gait recognition achieves competitive performance compared to video-based approaches in the cross-camera retrieval task and that gait and appearance features are complementary to each other. In addition, we compare gait recognition applied to walking and running sequences. An important difference is that we walk with straight arms, but run with bent arms. We propose to use human semantic parsing to create partial gait silhouettes from body parts to find the most discriminative combination. We demonstrate that the arm and leg swing are the most discriminative parts of the running gait. Our proposed method provides better recognition results by removing the torso from the silhouettes and allowing the arm swing to be more visible.

Although the pixel-wise labelling approaches have been exploited in depth and achieve good results in segmentation tasks, the grouped pixels are not ideal output for many end-users. In this paper, we propose a vertex-voting-based approach that can directly extract the polygon representations of objects. In order to better solve overlapping scenarios, we also propose a novel method that distinguishes objects by learning a virtual depth axis. When compared with the state-of-the-art method, our experiments demonstrate that this voting-based method is more robust to occlusion and shows a potential research direction.