The scalability of modern AI is fundamentally limited by the availability of labeled data. While supervised learning achieves remarkable performance, it relies on large annotated datasets, which are expensive and time-consuming to acquire. This work explores self-supervised learning (SSL) as a promising solution to this challenge, enabling AI to scale effectively in data-scarce scenarios. This study demonstrates the effectiveness of the proposed SSL framework using the EuroSAT dataset, a benchmark for land cover classification where labeled data is limited and costly. The proposed approach integrates contrastive learning with multi-spectral augmentations, such as spectral jittering and band... More >

Interpreting NMR spectra to accurately predict molecular structures remains a significant challenge in chemistry due to the complexity of spectral data and the need for precise structural elucidation. This study introduces NMRGen, a generative modeling framework that predicts molecular structures from NMR spectra and molecular formulas. The framework combines a SMILES autoencoder (GRU-based encoder-decoder) and an NMR encoder (CNN and DNN layers) to map spectral data to molecular representations. The SMILES autoencoder compresses and reconstructs SMILES strings, while the NMR encoder processes NMR spectra to generate latent vectors aligned with those from the SMILES encoder. Experiments were... More >

With the increasing global focus on renewable energy and the growing proportion of renewable power in the energy mix, accurate forecasting of renewable power demand has become crucial. This study addresses this challenge by proposing a multimodal information fusion approach that integrates time series data and textual data to leverage complementary information from heterogeneous sources. We develop a hybrid predictive model combining CNN and Bi-GRU architectures. First, time series data (e.g., historical power generation) and textual data (e.g., policy documents) are preprocessed through normalization and tokenization. Next, CNNs extract spatial features from both data modalities, which are... More >

Recently, the primary reason for blindness in adults has been diabetic retinopathy (DR) disease. Therefore, there is an increasing demand for a real-time efficient classification and detection system for diabetic retinopathy (DR) to overcome fast-growing disease (DR). We introduced a novel deep hybrid model for auto-mated diabetic retinopathy (DR) disease recognition and classification. Our model leverages the power of CNN architectures: Inception V3 and VGG16 models by combining their strengths to cater to exact requirements. VGG16 model efficiently captures fine features and wide-ranging features such as textures and edges, crucial for classifying initial signs of DR. Similarly, Inception... More >

This paper proposes an improved video action recognition method, primarily consisting of three key components. Firstly, in the data preprocessing stage, we developed multi-temporal scale video frame extraction and multi-spatial scale video cropping techniques to enhance content information and standardize input formats. Secondly, we propose a lightweight Inception-3D networks (LI3D) network structure for spatio-temporal feature extraction and design a soft-association feature aggregation module to improve the recognition accuracy of key actions in videos. Lastly, we employ a bidirectional LSTM network to contextualize the feature sequences extracted by LI3D, enhancing the representation capa... More >

In the era of rapid technological advancement, the demand for sophisticated Multi-Object Tracking (MOT) systems in applications such as intelligent surveillance and autonomous navigation has become increasingly critical.However, existing models often struggle with accuracy and efficiency in densely populated or dynamically complex environments. Addressing these challenges, we introduce a novel deep learning-based MOT model that incorporates the latest CT-DETR detection technology and an advanced ReID module for improved pedestrian tracking. Experimental results demonstrate the model's superior performance in accurately identifying and tracking multiple targets across varied scenarios, signif... More >

This paper proposes an improved object detection algorithm based on a dynamically deformable convolutional network (D-DCN), aiming to solve the multi-scale and variability challenges in object detection tasks. First, we review traditional methods in the field of object detection and introduce the current research status of improved methods based on multi-scale and variability convolutional neural networks. Then, we introduce in detail our proposed improved algorithms, including an improved feature pyramid network and a dynamically deformable network. In the improved feature pyramid network, we introduce a multi-scale feature fusion mechanism to better capture target information at different... More >

In this paper, we introduce a novel fast object detection framework, designed to meet the needs of real-time applications such as autonomous driving and robot navigation. Traditional processing methods often trade-off between accuracy and processing speed. To address this issue, we propose a hybrid data representation method that combines the computational efficiency of voxelization with the detail capture capability of direct data processing to optimize overall performance. Our detection framework comprises two main components: a Rapid Region Proposal Network (RPN) and a Refinement Detection Network (RefinerNet). The RPN is used to generate high-quality candidate regions, while the RefinerN... More >

With the rapid development of autonomous driving technology, the demand for real-time and efficient object detection systems has been increasing to ensure vehicles can accurately perceive and respond to the surrounding environment. Traditional object detection models often suffer from issues such as large parameter sizes and high computational resource consumption, limiting their applicability on edge devices. To address this issue, we propose a lightweight object detection model called YOLOv8-Lite, based on the YOLOv8 framework, and improved through various enhancements including the adoption of the FastDet structure, TFPN pyramid structure, and CBAM attention mechanism. These improvement... More >