It is necessary to locate microplastic particles mixed with beach sand to be able to separate them. This paper illustrates a kernel weight histogram-based analytical process to determine an appropriate neural network to perform tiny object segmentation on photos of sand with a few microplastic particles. U-net and MultiResUNet are explored as target networks. However, based on our observation of kernel weight histograms, visualized using TensorBoard, the initial encoder stages of U-net and MultiResUNet are useful for capturing small features, whereas the later encoder stages are not useful for capturing small features. Therefore, we derived reduced versions of U-net and MultiResUNet, such as Half U-net, Half MultiResUNet, and Quarter MultiResUNet. From the experiment, we observed that Half MultiResUNet displayed the best average recall-weighted F1 score (40%) and recall-weighted mIoU (26%) and Quarter MultiResUNet the second best average recall-weighted F1 score and recall-weighted mIoU for our microplastic dataset. They also require 1/5 or less floating point operations and 1/50 or a smaller number of parameters over U-net and MultiResUNet.

Multiscale forecasting of groundwater levels (GWLs) is essential for ensuring the sustainable management of groundwater resources, particularly considering the potential impacts of climate change. Such forecasting requires a model that is not only accurate in predicting GWLs but also computationally efficient, ensuring its suitability for practical applications. In this study, a temporal convolutional network (TCN) is implemented to forecast GWLs for 17 monitoring wells possessing diverse hydrogeological characteristics, located across South Korea. Using deep learning, the influence of meteorological variables (i.e., temperature, precipitation) on the forecasted GWLs was investigated by dividing the input features into three categories. Additionally, the models were developed for three forecast intervals (at 1-, 3-, and 6-month lead times) using each category input. When compared with state-of-the-art models, that is, long short-term memory (LSTM) and artificial neural network (ANN), the TCN model showed superior performance and required much less computational complexity. On average, the TCN model outperformed the LSTM model by 24%, 21%, and 25%, and the ANN model by 24%, 37%, and 47%, respectively, for 1-, 3-, and 6-month lead times. Based on these results, the proposed TCN model can be used for real-time GWL forecasting in hydrological applications.

We introduce a novel microscopic image dataset augmented with segmentation and detection labels specifically designed for microplastic analysis in sewage environments. Recognizing the increasing concern over microplastics—particles of synthetic polymers smaller than 5 mm—and their detrimental effects on marine ecosystems and human health, our research focuses on enhancing detection and analytical methodologies through advanced computer vision and deep learning techniques. The dataset comprises high-resolution microscopic images of microplastics collected from sewage, meticulously labeled for both segmentation and detection tasks, aiming to facilitate accurate and efficient identification and quantification of microplastic pollution. In addition to dataset development, we present example deep learning models optimized for segmentation and detection of microplastics within complex sewage samples. The models demonstrate significant potential in automating the analysis of microplastic contamination, offering a scalable solution to environmental monitoring challenges. Furthermore, we ensure the accessibility and reproducibility 12 of our research by making the dataset and model codes publicly available, accompanied by detailed 13 documentation on GitHub and LabelBox.

The Y-Maze behavioral test evaluates short-term spatial learning and memory in experimental mice, assessing the impact of various substances or conditions. Traditional methods require continuous manual monitoring, which is labor-intensive and prone to observer fatigue. To streamline this process, we propose an automated approach that not only tracks the critical positions of the mice but also monitors the layout of the Y-Maze itself. We explored the efficacy of conventional General Gaussian heatmaps alongside our novel Ellipse and Stadium heatmaps. Our findings indicate that our Stadium heatmap significantly enhances the precision of position recognition within the maze compared to traditional methods. Additionally, we address a common issue in heatmap regression—the Adaptive Wing loss, where the first derivative can become excessively high. By introducing and testing the Squared Adaptive Wing loss, we demonstrate its ability to improve the reliability of our automated system.

Keypoint localization is a fundamental task in computer vision, commonly addressed via heatmap regression. While the Soft-argmax Layer (SAM) exists as an end-to-end alternative, it suffers from fundamental limitations, including training instability and poor robustness, due to its internal softmax function. In this paper, we mathematically analyze these issues, identifying that the polarization characteristic of softmax causes information loss during forward propagation and a biased gradient flow during backpropagation. To resolve these problems, we propose the Linear-argmax Layer (LAM), a novel, softmax-free layer. LAM is a more robust and efficient architecture that ensures stable information flow through linear weighting. In experiments on face alignment datasets like WFLW and LAPA, LAM consistently achieved a lower prediction failure rate than SAM. Most notably, our lightweightLAMmodel with MobileNetV3-Small achieved a 5.39% NME with only 0.08 BFLOPs, overwhelmingly outperforming a heatmap-based model that recorded a 9.41% NME with over six times the computation. Consequently, our work demonstrates that LAM is a powerful alternative that overcomes the limitations of existing methodologies, enabling effective keypoint localization, especially in resource-constrained environments.