VTT Technical Research Centre of Finland

Abstract The primary aim is to develop a model that can achieve high accuracy in solving multi label classification problems by training, testing, and analyzing deep learning models that utilize both image and text data. In this paper, we propose a novel hybrid model that jointly considers textual and visual data for the task of movie genre prediction, which is a representative example of multi label classification problems. In the proposed model, textual features extracted from movie summaries are obtained using the DistilBERT model, while visual features derived from movie posters are extracted using the ConvNeXt deep learning model. These features from the two modalities are then combined using the XGBoost machine learning algorithm to perform genre prediction. This approach aims to achieve higher accuracy and better generalizability in movie genre classification by integrating information from different modalities through a late fusion method. The ConvNeXt architecture was adapted to the problem using transfer learning and fine-tuning techniques. To achieve the highest performance from the DistilBERT model, optimization was performed for the token length and threshold hyperparameters, and the model with a token length of 256 and a threshold of 0.5 was used in the hybrid model. Furthermore, to maximize the overall performance of the novel hybrid model, optimization was conducted using the Grid Search algorithm. All three proposed models were trained and tested on a dataset obtained from the IMDB website. The performances of the models were evaluated using hamming loss, precision and F1 score metrics. Experimental results revealed that, overall, the text-based model outperformed the image-based model, while the proposed hybrid model achieved higher performance than both individual models. It was demonstrated that textual and visual features complement each other and positively enhance the overall performance. This paper presents an original and effective study for multi label movie genre classification, combining the fields of computer vision, natural language processing, and machine learning.

Airborne and spaceborne sensors enable the rapid acquisition of large-area geospatial data, supporting multiple stages of mining operations, including exploration. While powerful, these techniques benefit from complementary proximal measurements to enhance mineralogical detail. This study investigates time-gated Raman spectroscopy as a field-deployable tool for Earth observation platforms, integrating a custom sampling solution for in situ analysis. Rock specimens and thin sections from Bockau, Germany, and Panagyurishte, Bulgaria, were examined under both field and laboratory conditions. Raman spectra revealed a minor systematic offset in peak positions compared with reference databases. Sixteen minerals were identified in the German samples and twenty-one in the Bulgarian samples, demonstrating the method's capability to deliver rapid, site-specific mineral characterization that complements remote sensing data.

The number of scrubber installations in the global ship fleet has increased in recent years as a response to tightened sulfur regulations set by the International Maritime Organization. In addition to mitigating sulfur oxides, scrubbers have been shown to provide some added benefits in terms of reducing particles in the exhaust. Recently, it was shown that treating the exhaust gas with a combination of a scrubber and a wet electrostatic precipitator can significantly reduce particle number, mass, and black carbon concentrations in the exhaust of marine engines. In this study, the effect of the system on the secondary aerosol formation potential was investigated. The exhaust aerosol produced during heavy fuel oil combustion readily formed aged particle mass, with the majority of the mass consisting of sulfate. The combination of the scrubber and WESP reduced the aged PM to levels corresponding to air utilized for dilution of the exhaust. For sulfate, the detected reduction efficiency of the system was more than 99%. The system was also effective in reducing primary PM during both heavy fuel oil and marine diesel oil combustion. Reduced primary and aged particle mass concentrations could indicate benefits in terms of PM-related health and visibility effects.

Lower critical solution temperature (LCST) behavior in water-soluble polymers arises from a delicate balance between polymer-water and polymer-polymer interactions, which in turn depend on backbone and side-chain structure. Recently, thionation (the replacement of the amide carbonyl oxygen by sulfur) was introduced as a new tool to adjust LCST behavior in poly(2-oxazoline)s by weakening hydrogen bonding and reducing polarity. However, it remains unclear how this approach translates across different amide-based polymers and LCST mechanisms.

This study focuses on a design–build (DB) procedure that exploits the related parties’ (preconstruction) collaboration to improve the economic efficiency of construction projects. It is called design–build with a development phase (DBd). In the procedure, the owner and the selected contractor continue the development of the project solution in cooperation, adhering to the principle of benefit sharing, which is enabled by preagreed rules and the benchmark solution, formed as a result of requirement-based price-inclusive contractor selection where technical design is not included in the proposal; the formation of the final contract is a two-stage process, and a new owner decision is needed to enter the realization phase. More precisely, this study aims to discover and formulate a theory that underlies the DBd procedure that is a heuristic construct of the procedure’s origin. For that, the DBd procedure was conceptualized by determining its key elements and their relationships, and by showing how a coherent whole is achieved by the elements’ mutual interoperability. Therefore, the study contributes to the body of knowledge by crystallizing the DBd concept and its rationality. Even more importantly, the study eventually establishes the novel DBd procedure by combining consistent theoretical and practical views considering the earlier success of DBd in trial projects.

Sheets made fromlignin-rich fiber raw materials can be bondedby hot pressing without external binders. This paper explores howair-laid, foam-laid, and water-laid web formation methods, initialsheet moisture content, as well as hot-pressing conditions (5 MPa,100–260 °C, 1–60 s), impact the physical propertiesof board-like materials made of chemi-thermomechanical softwood fibers.In addition to the structural characterization of the hot-pressedmaterials by X-ray microtomography, air permeance, water contact angle,dry and wet tensile strength, and in-plane compression propertieswere measured. Despite the significant structural densification, characteristicsof the forming method were retained after hot pressing in the finalsheet properties. The compressed air-laid sheets had the highest airpermeance and the smallest mean pore size, which could be beneficialfor particle filtering. At moderate pressing temperatures and times,the significant proportion of large pores in the foam-laid sheetsmade them weaker than the corresponding water-laid sheets. However,under extreme pressing conditions, the foam- and water-laid sheetsreached similar values of high tensile and in-plane compression strength.This suggests that polymer interdiffusion becomes the dominant factorfor material strength under these conditions, superimposing the hydrogenbonding created during aqueous forming.

This summary is going to deal with those papers and poster session presentations which may be covered by the heading: Frost action in soil. These papers were mainly presented in session 1. The space is limited and therefore it is possible to mention only those points which the authors feel to be most important.

BackgroundRespiratory rate is an important part of assessing the clinical state of children, and various methods exist to measure it. However, there is a lack of a universally accepted reference standard to validate the performance of these methods.AimTo identify different reference standards that have been used to evaluate respiratory rate measurement methods in children under 5 years of age and describe their perceived strengths and limitations.MethodsMEDLINE and Web of Science were searched for studies in English. Studies of children under 5 years of age, published between 2013 and 2024, in which a method for measuring respiratory rate was compared against a reference standard, were included. Deductive content analysis was used to map perceived strengths and limitations of each standard, and a forest plot analysis was used to compare agreement between the reference standard and the index tests.ResultsFrom 992 retrieved studies, 56 were included. The most common reference standard was impedance pneumography (22/56), primarily used in high-income settings, followed by manual counting (19/56), mostly employed in low- and middle-income settings, and capnography (9/56). Child age, clinical condition, setting, training of personnel and the ease of implementation were all important factors in which the reference standard was used and how it performed.ConclusionThree different reference standards were used for most studies; however, their relative performance to each other is unclear. There is a need for research that directly compares the performance of these reference standards across different age strata and settings in order to confidently recommend a reference standard for respiratory rate measurement methods.