Reliable condition monitoring of transmission line fittings is essential to improve grid safety and reduce maintenance costs. However, defect detection in unmanned aerial vehicle (UAV) inspection imagery is challenging due to extreme scale variation, tiny targets such as bolts and pins that are prone to false positives and missed detections, cluttered backgrounds, and hard samples such as small objects interspersed among regular-sized objects that slow convergence. To address these issues, this study proposes an artificial intelligence-based defect detection method that combines a two-stage transfer learning strategy with image background denoising, enabling the detector to gradually learn discriminative features of transmission line fittings. On this basis, the You Only Look Once version 11 (YOLOv11) detector is improved to enhance multi-scale feature extraction, improve adaptability to small targets, and mitigate the influence of hard samples during training. Experimental results show that the proposed method achieves a precision of 88.4%, a mean Average Precision of 81.3% at an Intersection over Union threshold of 0.5 ([email protected]), and a mean Average Precision of 60.3% averaged over Intersection over Union thresholds from 0.5 to 0.95 ([email protected]:0.95), while running at 68.1 frames per second (FPS). Compared with the corresponding baseline detector, the proposed method improves these metrics by 4.7, 4.5, and 4.9 percentage points, confirming its effectiveness for transmission line fitting defect detection in complex backgrounds and its practical potential for near-real-time UAV-assisted transmission line inspection.

As the number of Internet of Things (IoT) devices continues to surge, particularly in remote areas, ensuring efficient and reliable connectivity has become a critical challenge. The integration of Narrowband IoT (NB-IoT) with Non-Terrestrial Networks (NTN) offers a promising solution. However, in Low Earth Orbit (LEO) satellite environments, existing random access (RA) procedures face significant issues, especially with high user density, leading to increased collisions and reduced access efficiency. While progress has been made in areas like delay compensation and Doppler effect management, challenges in collision mitigation remain. This paper proposes two innovative strategies to enhance RA performance in NB-IoT NTN systems: a non-orthogonal frequency hopping pattern for the narrowband physical random access channel (NPRACH) preamble, which increases the number of available preambles by allowing controlled overlap in frequency hopping sequences, thus reducing collision rates; and an adaptive access mechanism, which dynamically adjusts access parameters like backoff intervals and access barring factors based on network conditions to improve access success rates. Simulation results demonstrate the effectiveness of these improvements, especially in high-traffic environments, providing a more efficient and reliable solution for future NB-IoT deployments in NTN.

Traditionally, agricultural forecasting has relied on empirical methods and basic statistical analysis, such as applying average values from previous years’ yields or using a simple linear fit for next year’s predictions. However, the emergence of data-driven approaches, particularly machine learning algorithms, has revolutionized yield prediction in agriculture. Machine learning techniques have demonstrated their potential to provide accurate predictions. However, existing models often rely on a limited number of input variables for crop yield predictions, which makes them only suitable for specific scenarios. In this study, we have developed four distinct machine learning-based predictors, incorporating various climate factors, including daytime temperature, nighttime temperature, precipitation (rainfall), vegetation index, and evapotranspiration as input variables to predict tomato acreage yields in counties of California, USA. Our results show that regression models constructed using neural networks and linear regression exhibited better performance than other predictors, achieving an average accuracy rate of 70% to 80%. Compared to most of the existing crop yield predictors, our models offer versatility while maintaining a desirable level of predictive accuracy. Expanding the number of input variables, such as nitrogen fertilizer usage etc, and introducing larger spatial and temporal high-resolution datasets for model training can improve our model performance, enabling us to obtain better results in tomato yield prediction.

The Non-Autoregressive Transformer, due to its low inference latency, has attracted much attention from researchers. Although, the performance of the non-autoregressive transformer has been significantly improved in recent years, there is still a gap between the non-autoregressive transformer and the autoregressive transformer. Considering the success of localness on the autoregressive transformer, in this work, we consider incorporating localness into the non-autoregressive transformer. Specifically, we design a dynamic mask matrix according to the query tokens, key tokens, and relative distance, and unify the localness module for self-attention and cross-attention module. We conduct experiments on several benchmark tasks, and the results show that our model can significantly improve the performance of the non-autoregressive transformer.

Current spatio-temporal action detection methods usually employ a two-stream architecture, a RGB stream for raw images and an auxiliary motion stream for optical flow. Training is required individually for each stream and more efforts are necessary to improve the precision of RGB stream. To this end, a single stream network named enhanced action tubelet (EAT) detector is proposed in this work based on RGB stream. A modulation layer is designed to modulate RGB features with conditional information from the visual clues of optical flow and human pose. This network is end-to-end and the proposed layer can be easily applied into other action detectors. Experiments show that EAT detector outperforms traditional RGB stream and is competitive to existing two-stream methods while free from the trouble of training streams separately. By being embedded in a new three-stream architecture, the resulting three-stream EAT detector achieves impressive performances among the best competitors on UCF-Sports, JHMDB and UCF-101.

With the growth of urban population, crowd analysis has become an important and necessary task in the field of computer vision. The goal of crowd counting, which is a subfield of crowd analysis, is to count the number of people in an image or a zone of a picture. Due to the problems like heavy occlusions, perspective and luminous intensity variations, it is still extremely challenging to achieve crowd counting. Recent state-of-the-art approaches are mainly designed with convolutional neural networks to generate density maps. In this work, Multi-Dilation Network (MDNet) is proposed to solve the problem of crowd counting in congested scenes. The MDNet is made up of two parts: a VGG-16 based front end for feature extraction and a back end containing multi-dilation blocks to generate density maps. Especially, a multi-dilation block has four branches which are used to collect features in different sizes. By using dilated convolutional operations, the multi-dilation block could obtain various features while the maximum kernel size is still 3 x 3. The experiments on two challenging crowd counting datasets, UCF_CC_50 and ShanghaiTech, have shown that the proposed MDNet achieves better performances than other state-of-the-art methods, with a lower mean absolute error and mean squared error. Comparing to the network with multi-scale blocks which adopt larger kernels to extract features, MDNet still gains competitive performances with fewer model parameters.