Search

Efficient management of agricultural resources de-mands a profound comprehension of plant growth dynamics, focusing on sustainable methodologies. This study delves into the forestry sector in Sweden, explicitly addressing the critical early phases of pine tree development within controlled en-vironments, predominantly nurseries and specialised facilities. To tackle the complexities of temporal classification in pine tree growth, we propose a novel measurement system that uses image data to leverage the power of different deep transfer learning-based convolutional neural networks. To classify pine trees over time, our approach integrates different state-of-the-art transfer learning models based on the features extracted by SqueezeNet, MobileNetV2, GoogLeNet, ShuffieNet, and ResNetXt into our measurement system. The primary challenge in tem-porally sorting plant growth is that instances within the same class exhibit variations, whereas instances belonging to different classes may share similarities. We also addressed a pivotal question: whether focusing on the region of interest (ROI) as the input improves the final results. To explore this, we defined two distinct scenarios and conducted a comparative analysis to elucidate the impact of varied input images on the efficacy of our measurement systems. Our experimental results highlight the superior performance of deep SqueezeNet transfer learning over other models, with an error reduction of 15% compared to the second-best approach. Additionally, we provide a sensitivity analysis of the hyperparameters.

The efficient management of agricultural resources requires a deep understanding of plant growth dynamics. This research focuses on Sweden's forestry sector and explicitly addresses the crucial early stages of pine tree development. The main difficulty in categorising plant growth over time is that instances within a given category are not identical, while instances from different categories may have similarities. In this context, we present a novel measurement system that integrates the capa-bilities of evolutionary computation and deep transfer learning using image data. The image acquisition system includes a tray of plates that moves through a nursery, generating a dataset captured over 44 days of plant growth. Our newly proposed algorithm, EvoSqueezeNet, employs various search strategies for SqueezeNet deep transfer learning to find proper hyper-parameters. We opted for SqueezeNet based on our preliminary studies, revealing its superior performance compared to other pre-trained models in our case study. Given that our approach is independent of any specific evolutionary algorithm, we utilised five distinct search strategies. These include Differential Evolution (DE), Particle Swarm Optimisation (PSO), Covariance Ma-trix Adaptation Evolution Strategy (CMA-ES), Comprehensive Learning PSO (CLPSO), and Linear Population Size Reduction Success-History Adaptation DE (LSHADE). Consequently, we proposed five EvoSqueezeNet schemes for temporal plant growth categorisation. One characteristic of our proposed model is that it uses a limited computation budget for search strategies, en-hancing its applicability in real-world applications. The proposed EvoSqueezeNet methodology demonstrates an error reduction of more than 40%, showcasing its superior performance compared to competing methods.

Eliminating Multi-Path Interference (MPI) stands as a significant unresolved challenge in the domain of depth estimation using Time-of-Flight (ToF) cameras. ToF data is typically influenced by significant noise and artifacts stemming from MPI. Although a variety of conventional methods have been suggested to enhance ToF data quality, the application of machine learning techniques has been infrequent, primarily due to the scarcity of authentic training data with accurate depth information. This paper introduces an approach that eliminates the dependency on labeled real-world data within the learning framework. We employ a U-Net trained on the data with ground truth in a supervised manner, enabling it to leverage multi-frequency ToF data for MPI correction. Concurrently, we compare three channels as input with one channel and two channels. Our experimental results convincingly showcase the effectiveness of this approach in reducing noise in real-world data.

Snow accumulations, especially if of great intensity, as is the case in northern countries, for example, can be very damaging, especially if they occur in urban environments. The damage provoked by snow is not only related to the weight of the accumulations, causing damage to structures but also to the pollution retained by the structure of the snowflakes. However, snow weight monitoring is a complex task, both because of the placement of the sensors and the specific operating ranges they must have in terms of operating temperature. These complications can be overcome by the design and use of a soft sensor, that is, a sensor capable of making indirect measurements from other parameters related to the measurement under consideration. This paper presents the design and metrological validation of a soft sensor for indirect weight measurement of snow accumulations. The designed soft sensor has been based on Artificial Neural Network and achieved, as a result, a Root-Mean-Square Error (RMSE) of 114g and a maximum extended uncertainty, evaluated by Monte Carlo simulation, of 300g in a measurement range from 150g to 5200g.

The development of machine learning has resulted in data gaining a pivotal role in the technological advancement, especially data where the ground truth of targeted parameters can be efficiently captured. This requires the development of methods that facilitate accurate data collection with ground truth. Under this perspective, Time of Flight sensors pose a high complexity due to the multifaceted nature of noise in the captured data. To enable the use of such sensors in a wide range of applications including Artificial Intelligence, we need to provide also accurate ground truth data. In this article, we present a method for automated data capturing from a LiDAR sensor together with ground truth data generation. This method will facilitate generating big datasets from LiDAR sensors with high accuracy ground truth data. In addition, we provide a dataset that aside from depth sensor data contains also RGB, confidence and infrared data captured from the LiDAR sensor. As a result, the proposed method not only facilitates data capturing but it enables to generate accurate ground truth data, with RMSE of only 0.04 m at 1.3 m distance.

As we find ourselves in the midst of the fourth industrial revolution, also known as Industry 4.0, the digital transformation of products, processes, and systems, along with their interconnectedness, is of utmost interest. To ensure future competitiveness in the manufacturing sector, the integration of advanced manufacturing technologies and advanced information technology is essential. Information technologies and knowledge are deeply intertwined with industrial equipment, processes, products, and systems, posing a challenge in transitioning today's manufacturing industry into the digital era. The manufacturing sector will require adequate methods, a conducive working environment, new tools, and lifelong training to support its employees. This article describes a joint effort of five Swedish universities with the ambition to strengthen the competitiveness and innovativeness of the national manufacturing industry through highly competent researchers and future leaders. The collaboration is in the form of an industrial graduate school, combining the efforts of five universities, 16 graduate students, and 12 companies or organisations. This article will outline how the graduate school has been organized, the joint efforts that have been made to assure the development of all parties, organisations and individuals, and will also outline some of the key success factors that have been identified thus far in the project., CC BY-NC 4.0 DEEDCorresponding Author: Peter Thorvald, peter.thorvald@his.se, Smart Industry Sweden

I dagens samhälle blir det allt viktigare att fortbilda sig under hela sitt yrkesverksamma liv. För att möta efterfrågan på det livslånga lärandet har Mittuniversitetet utvecklat och genomfört ett antal kurser som riktar sig mot yrkesverksamma ingenjörer. Detta arbete presenterar våra erfarenheter av att ge dessa kurser, med en tyngdpunkt på studenternas upplevelser. Syftet med detta är att bygga upp en vetenskaplig bas för vad vi gör som är bra, men även vad som kan förbättras och förändras. Målsättningen är att göra dessa specialanpassade kurser riktade mot yrkesverksamma ingenjörer så givande och flexibla som möjligt. Våra initiala resultat visar bland annat att studenternas negativa upplevelser ofta var kopplade till antagningsförfarandet och det praktiska genomförandet av kurserna. Man hade svårigheter med att hitta hur man skulle registrera sig på kursen och att tidsramen för registrering kunde vara ett problem. Läroplattformen uppfattades som svår att överblicka och det förekom även viss otydlighet gällande var undervisningen skulle äga rum. Den positiva responsen i utvärderingarna gällde främst det faktiska kursinnehållet, då man ansåg att uppgifter och kursmaterial var givande. Vidare uppskattades kursupplägget, att man kunde kombinera studierna med arbete. Framledes kommer vi att fortsätta med dessa utvärderingar i takt med att kurserna ges, och därefter anpassa vårt mottagande och kommunikationen med studenterna. Även kursupplägget ses över kontinuerligt via den återkoppling vi mottar.

Implementing Convolutional Neural Networks (CNN) based computer vision algorithms in Internet of Things (IoT) sensor nodes can be difficult due to strict computational, memory, and latency constraints. To address these challenges, researchers have utilized techniques such as quantization, pruning, and model partitioning. Partitioning the CNN reduces the computational burden on an individual node, but the overall system computational load remains constant. Additionally, communication energy is also incurred. To understand the effect of partitioning and pruning on energy and latency, we conducted a case study using a feet detection application realized with Tiny Yolo-v3 on a 12th Gen Intel CPU with NVIDIA GeForce RTX 3090 GPU. After partitioning the CNN between the sequential layers, we apply quantization, pruning, and compression and study the effects on energy and latency. We analyze the extent to which computational tasks, data, and latency can be reduced while maintaining a high level of accuracy. After achieving this reduction, we offloaded the remaining partitioned model to the edge node. We found that over 90% computation reduction and over 99% data transmission reduction are possible while maintaining mean average precision above 95%. This results in up to 17x energy savings and up to 5.2x performance speed-up.

Detecting and measuring objects with vision-based systems in uncontrolled environments is a difficult task that today, thanks to the development of increasingly advanced artificial intelligence-based techniques, can be solved with greater ease. In this context, this article proposes a novel approach for the vision-based measurement of objects in uncontrolled environments using a specific type of convolutional neural network (CNN) named you only look once (YOLO) and a direct linear transformation (DLT) process. The case study concerned designing a novel vision-based system for measuring the diameter of wood logs cut and loaded onto trucks. This problem has been occurring in the Swedish forestry industry. In fact, this operation is not carried out with computer vision algorithms because of the high variability of environmental conditions caused by the changing position of the sun, weather conditions, and the variability of truck positioning. To solve this problem, the YOLO network is proposed to locate logs while attempting to maintain a high Intersection over Union (IoU) value for the correct estimation of log size. Furthermore, in order to obtain accurate measurements, the DLT is used to convert into world coordinates the dimensions of the logs themselves. The proposed CNN-based solution is described after briefly introducing today’s methodologies adopted for wood bundle analysis. Particular attention is paid to both the training and the calibration steps. Results report that for 80% of cases, the error reported has been smaller than 4 cm, representing only 8% of the measurement, considering a mean log diameter for the application of 50 cm.

Computer vision systems in sensor nodes of the Internet of Things (IoT) based on Deep Learning (DL) are demanding because the DL models are memory and computation hungry while the nodes often come with tight constraints on energy, latency, and memory. Consequently, work has been done to reduce the model size or distribute part of the work to other nodes. However, then the question arises how these approaches impact the energy consumption at the node and the inference time of the system. In this work, we perform a case study to explore the impact of partitioning a Convolutional Neural Network (CNN) such that one part is implemented on the IoT node, while the rest is implemented on an edge device. The goal is to explore how the choice of partition point, quantization method and communication technology affects the IoT system. We identify possible partitioning points between layers, where we transform the feature maps passed between layers by applying quantization and compression to reduce the data sent over the communication channel between the two partitions in Tiny YOLOv3. The results show that a reduction of transmitted data by 99.8% reduces the network accuracy by 3 percentage points. Furthermore, the evaluation of various IoT communication protocols shows that the quantization of data facilitates CNN network partitioning with significant reduction of overall latency and node energy consumption.

Anomaly detection in multivariate time series is valuable for many applications. In this context, unsupervised and semi-supervised deep learning methods that estimate how normal a new observation is have shown promising results on benchmark datasets. These methods are dependent on a threshold that determines which points should be regarded as anomalous and not be anomalous. However, finding the optimal threshold is not easy since no information about the ground truth is known in advance, which implies that there are limitations to automatic threshold-setting methods available today. An alternative is to utilize the expertise of users that can interact in a threshold-setting procedure, but for this to be practically feasible, the method needs to be both accurate and efficient in relation to the state-of-the-art automatic methods. Therefore, this study develops an interactive threshold-setting schema and examines to what extent it can outperform the current state-of-the-art automatic threshold-setting methods. The result of the study strongly indicates that the suggested method with little effort can provide higher accuracy than the automatic threshold-setting methods on a general basis.

In a time when Artificial Intelligence (AI) technologies are nearly ubiquitous, railway construction and maintenance systems have not fully grasped the capabilities of such technologies. Traditional railway inspection methods rely on inspection from experienced workers, making such tasks costly from both, the monetary and the time perspective. From an overview of the state-of-the-art research in this area regarding AI-based systems, we observed that their main focus was solely on detection accuracy of different railway components. However, if we consider the critical importance of railway fastening in the overall safety of the railway, there is a need for a thorough analysis of these AI-based methodologies, to define their uncertainty also from a metrological perspective. In this article we address this issue, proposing an image-based system that detects the rotational displacement of the fastened railway clips. Furthermore, we provide an uncertainty analysis of the measurement system, where the resulting uncertainty is of 0.42°, within the 3° error margin defined by the clip manufacturer.

The importance of preventing failures in bearings has led to a large amount of research being conducted to find methods for fault diagnostics and prognostics. Many of these solutions, such as deep learning methods, require a significant amount of data to perform well. This is a reason why publicly available data are important, and there currently exist several open datasets that contain different conditions and faults. However, one challenge is that almost all of these data come from a laboratory setting, where conditions might differ from those found in an industrial environment where the methods are intended to be used. This also means that there may be characteristics of the industrial data that are important to take into account. Therefore, this study describes a completely new dataset for bearing faults from a pulp mill. The analysis of the data shows that the faults vary significantly in terms of fault development, rotation speed, and the amplitude of the vibration signal. It also suggests that methods built for this environment need to consider that no historical examples of faults in the target domain exist and that external events can occur that are not related to any condition of the bearing.

Across all industries, digitalization and automation are on the rise under the Industry 4.0 vision, and the forest industry is no exception. The forest industry depends on distributed flows of raw materials to the industry through various phases, wherein the typical workflow of timber loading and offloading is finding traction in using automation and 5G wireless networking technologies to enhance efficiency and reduce cost. This article presents one such ongoing effort in Sweden, Remote-Timber—demonstrating a 5G-connected teleoperation use-case within a workflow of timber terminal—and disseminates its business attractiveness as well as first measurement results on network performance. Also, it outlines the future needs of the 5G network design/optimization from teleoperation perspective. Overall, the motivation of this article is to disseminate our early-stage findings and reflections to the industrial and academic communities for furthering the research and development activities in enhancing 5G networks for verticals.

Significant risks to public health and the environment are posed by the release of hazardous gases from industries such as pulp and paper. In this study, the aim was to develop a multi-sensor system with a minimal number of sensors to detect and identify hazardous gases. Training and test data for two gases, hydrogen sulfide and methyl mercaptan, which are known to contribute significantly to odors, were generated in a controlled laboratory environment. The performance of two deep learning models, a 1d-CNN and a stacked LSTM, for data fusion with different sensor configurations was evaluated. The performance of these models was compared with a baseline machine learning model. It was observed that the baseline model was outperformed by the deep learning models and achieved good accuracy with a four-sensor configuration. The potential of a cost-effective multi-sensor system and deep learning models in detecting and identifying hazardous gases is demonstrated by this study, which can be used to collect data from multiple locations and help guide the development of in-situ measurement systems for real-time detection and identification of hazardous gases at industrial sites. The proposed system has important implications for reducing pollution and protecting public health.

The widespread adoption of Deep Learning techniques for Computer Vision in recent years has brought major changes to the world of industry, contributing greatly to this sector's transition to Industry 4.0, also referred to as Smart Industry. This involves an increasingly predominant role of machines and automation within industrial processes. In this context, the Swedish forest industry is an excellent context for applying these techniques. In particular, this work will deal with automating the measurement of log diameters to date carried out manually by operators in the industry. The proposed methodology will use two object detection neural networks, one deputed to detect logs in the scene and the other for the calibrated target. The latter thus allows the camera calibration to be fully automated, enabling each diameter to be measured without any further operations by the operator. The results obtained are satisfactory and open the way for the industrial application of the proposed methodology.

Skin cancers are the most cancers diagnosed worldwide, with an estimated > 1.5 million new cases in 2020. Use of computer-aided diagnosis (CAD) systems for early detection and classification of skin lesions helps reduce skin cancer mortality rates. Inspired by the success of the transformer network in natural language processing (NLP) and the deep convolutional neural network (DCNN) in computer vision, we propose an end-to-end CNN transformer hybrid model with a focal loss (FL) function to classify skin lesion images. First, the CNN extracts low-level, local feature maps from the dermoscopic images. In the second stage, the vision transformer (ViT) globally models these features, then extracts abstract and high-level semantic information, and finally sends this to the multi-layer perceptron (MLP) head for classification. Based on an evaluation of three different loss functions, the FL-based algorithm is aimed to improve the extreme class imbalance that exists in the International Skin Imaging Collaboration (ISIC) 2018 dataset. The experimental analysis demonstrates that impressive results of skin lesion classification are achieved by employing the hybrid model and FL strategy, which shows significantly high performance and outperforms the existing work.

To produce flawless glass containers, continuous monitoring of the glass gob is required. It is essential to ensure production of molten glass gobs with the right shape, temperature, viscosity and weight. At present, manual monitoring is common practice in the glass container industry, which heavily depends on previous experience, operator knowledge and trial and error. This results in inconsistent measurements and consequently loss of production. In this article, a multi-camera based setup is used as a non-invasive real-time monitoring system. We have shown that under certain conditions, such as keeping the glass composition constant, it is possible to do in-line measurement of viscosity using sensor fusion to correlate the rate of geometrical change in the gob and its temperature. The correlation models presented in this article show that there is a strong correlation, i.e., 0.65, between our measurements and the projected viscosity.

The sophistication and high accuracy of Deep Neural Networks have gotten significant attention in recent years, with a wide range of applications making use of their capabilities. However, the deployment of such networks still faces limitations due to the high volume of data to be processed and the high computational requirements. In this article we focus on the effects that data volume reduction, due to image compression and scaling down the image resolution, will have on the detection accuracy for the design case of a powered wheelchair guidance system. Throughout our analysis we show that the reduction in image resolution to a factor of 16× in image area alongside with JPEG compression provides a detection accuracy of over 0.93 in mAP, while the additional error in the position estimation of the caregiver is less than 0.5 cm. By reducing the data volume we inherently reduce the communication energy consumption, which is reduced by more than one order of magnitude. These results prove that we can overcome the complexity of high data volume for the deployment of DNNs in resource constrained IoT applications by interlacing the effects of image compression and resolution reduction, maintaining the accuracy and reducing the node energy consumption.

The importance of anomaly detection in multivariate time series has led to the development of several prominent deep learning solutions. As a part of the anomaly detection method, the scoring method has shown to be of significant importance when separating non-anomalous points from anomalous ones. At this time, most of the solutions utilize an aggregated score which means that relevant information created by the anomaly detection model might be lost. Therefore, this study has set out to examine to what extent anomaly detection in multivariate time series based on deep learning can be improved if all the residuals from each individual channel is considered in the anomaly score. To achieve this, an aggregated and separated scoring method has been applied with a simple denoising convulutional autoencoder (DCAE). In addition, the performance has been compared with other state-of-the-art methods. The result showed that the separated approach has the potential to generate a significantly higher performance than the aggregated one. At the same time, there were some indications suggesting that an aggregated scoring is better at generalizing when no labels to base the anomaly thresholds on, are available. Therefore, the result should serve as an encouragement to use a separated scoring approach together with a small sample of labeled anomalies to optimise the thresholds. Lastly, due to the impact of the anomaly score, the result suggests that future research within this field should consider applying the same anomaly scoring method when comparing the performance of deep learning algorithms.

The study aims to explore the fuel consumption of city buses with data fusion using a dataset with multiple parameters such as travelled distance, weekday, hour of the day, drivers, buses, and routes, that influence the trip fuel consumption. In this study, manipulated parameters such as modified driver, bus and route identification numbers are used together with original parameters to identify the optimal combination of parameters that can be used to enhance the accuracy of the prediction model. Two regression methods, i.e. cubic SVM and artificial neural networks (ANN), are used to demonstrate the performance of the proposed approach. Results shows that a combination of original parameters and processed parameters increases the performance.

Skin cancer is one of the most threatening cancers, which spreads to the other parts of the body if not caught and treated early. During the last few years, the integration of deep learning into skin cancer has been a milestone in health care, and dermoscopic images are right at the center of this revolution. This review study focuses on the state-of-the-art automatic diagnosis of skin cancer from dermoscopic images based on deep learning. This work thoroughly explores the existing deep learning and its application in diagnosing dermoscopic images. This study aims to present and summarize the latest methodology in melanoma classification and the techniques to improve this. We discuss advancements in deep learning-based solutions to diagnose skin cancer, along with some challenges and future opportunities to strengthen these automatic systems to support dermatologists and enhance their ability to diagnose skin cancer. Author

Since early-stage skin cancer identification can improve melanoma prognosis and significantly reduce treatment costs, AI-based diagnosis systems might greatly benefit patients suffering from suspicious skin lesions. The study proposes a cosine cyclical learning rate with a skin cancer classification model to improve melanoma prediction. The contributions of models involve three critical CNNs, which are standard deep feature extraction modules for the skin cancer classification in this study (Vgg19, ResNet101 and InceptionV3). Each CNN model applies three different learning rates: fixed learning rate(LR), Cosine Annealing LR, and Cosine Annealing with WarmRestarts. HAM10000 is a large collection of publicly available dermoscopic images dataset used for our experiments. The performance of the proposed approach was appraised through comparative experiments. The outcome has indicated that the proposed method has high efficiency in diagnosing skin lesions with a cosine cyclical learning rate.