ABSTRACT Remote sensing image target recognition plays an important role in environmental monitoring, disaster management, urban planning, and traffic systems. To address issues such as high missing detection rates of dense small targets and insufficient multi‐scale feature interaction in remote sensing image target detection, an improved PAP‐YOLOv8 network architecture was proposed in this study based on YOLOv8. This framework included three core innovations: (1) the pixel dynamic convolution attention module (PDCA), which enhanced small target detection capabilities in complex backgrounds through direction‐sensitive convolution and a dual‐stage attention mechanism; (2) the adaptive channel attention network (ACAN), which optimized multi‐scale feature fusion and suppressed background interference via dynamic weight allocation and channel recalibration; (3) the pyramidal wavelet attention fusion block (PWAFB), which synergistically integrated wavelet multi‐resolution analysis with composite attention mechanisms to model spatial details and channel relationships in remote sensing images through a collaborative decomposition and reconstruction workflow. Experiments on the Detection in Optical Remote Sensing (DIOR) dataset showed that the [email protected] of PAP‐YOLOv8 reached 86.9%, an increase of 5.3% over the baseline YOLOv8, while demonstrating enhanced detection capability for small and multi‐scale targets. Compared with other detection models, PAP‐YOLOv8 achieves a balanced trade‐off between network size and detection capability, demonstrating superior overall performance with an F1 score of 83.1%. This study confirms the potential of PAP‐YOLOv8 for satellite monitoring applications and provides a valuable reference for improving target detection performance in optical remote sensing satellite imagery.

ABSTRACT The technology for target recognition in remote sensing satellite images is widely applied in daily life, and research on detecting and recognizing targets in remote sensing images holds significant academic and practical importance. To address the challenges of extreme scale variations, dense target distributions, and low‐resolution artefacts in remote sensing images, this paper proposes a new multi‐object detection network based on the YOLOv8 architecture—MD‐YOLOv8. The main contributions of this paper are threefold: (1) the design of the multi‐frequency attention downsampling module, which integrates the ADown module with Haar wavelet transforms and pixel attention; (2) the proposal of the adaptive attention network (DMAA) module, an enhanced multiscale feature extractor based on the multiscale feature extraction attention mechanism; (3) the integration of both modules into the YOLOv8 backbone to achieve superior performance in remote sensing image detection. Based on the DOTA‐1.0 dataset for training, experimental results show that the MD‐YOLOv8 network achieves improvements in precision, recall rate, and [email protected], reaching 82.69%, 78.28%, and 82.05%, respectively; these represent increases of 3.76%, 3.43%, and 4.37% compared to the original model. In practical image detection, MD‐YOLOv8 demonstrates higher recognition quality and can flexibly respond to various target types. The MD‐YOLOv8 network effectively meets the accuracy requirements for target detection in remote sensing satellite images.

The growing complexity of modern Internet applications and the widespread use of microservice architectures have amplified the need for efficient trace anomaly detection to maintain system stability. Despite the fact that many trace anomaly detection algorithms have been proposed to identify abnormal behaviors, a comprehensive evaluation of these methods is lacking, which makes it difficult for developers to choose the most suitable algorithm for real-world applications. To address this gap, we presentTADBench, a comprehensive and extensible benchmark for trace anomaly detection.TADBenchconsolidates diverse publicly available trace datasets and algorithms into a unified repository, standardizes data formats, and incorporates manual anomaly labels. To ensure reproducibility and fair comparisons, we propose a modular evaluation framework supporting end-to-end model assessment. Additionally, we provide practical guidance for algorithm selection based on specific data attributes by evaluating their performance across datasets with different characteristics, thereby effectively bridging the gap between academic research and industrial deployment. To the best of our knowledge, this is the first comprehensive empirical study of trace anomaly detection algorithms. Our findings aim to facilitate the adoption of these methods in production environments, offering actionable insights for developers and researchers.

The recent success of audio language models (LMs) has revolutionized the field of neural music generation. Among all audio LM approaches, MusicGen has demonstrated the success of a single-stage LMs based music generation framework, without needing to train multiple LMs. Despite its promising performance in generating monophonic (mono) music, directly generating stereophonic (stereo) music following the previous framework has resulted in perceptible quality degradation. In this paper, we first discuss the difficulty of directly encoding stereo music with neural codec, and then provide a stable and practical solution based on a dual encoding approach. To utilize the dually encoded tokens in single-stage LMs, we also propose two forms of token sequence patterns. An extensive evaluation has been conducted using various aspects of stereo music audios to examine the performance of stereo neural codec approaches and the generation quality of single-stage LMs. Finally, our experimental results suggest that (i) our proposed dual encoding approach for neural codec is significantly better than the typical joint encoding approach in terms of reconstruction quality, and (ii) the stereo single-stage LMs trained with our proposed token sequence patterns substantially improved the perceptual quality of the state-of-the-art music generation model (i.e. MusicGen) in subjective tests.