A novel indoor positioning method for a pigsty based on ultra-wideband (UWB) technology is proposed via reasonably arranging the base stations and positioning tags, which can provide strong support for the accurate positioning of specific pig pens in the pig house. The distribution of Dilution of Precision (DOP) values is analyzed in different deployment schemes which include the amount, the height, the deployment shape, and the distance of the base stations through positioning simulation. Moreover, the positioning effect of dynamic following positioning, the root-mean-square error (RMSE), and the variance of static positioning are measured. It is noteworthy that the Kalman filtering algorithm is adopted to smooth the dynamic following positioning experimental data for reducing the positioning error. The simulated results show that the DOP values decrease with the increase in height and distance, and these results and laws are consistent with the experimental results. However, the DOP values are best reached when the height is from 1.75 m to 2.75 m, and the distance is over 3.00 m. Furthermore, the DOP values of quadrilateral deployment are smaller than the ones of triangular deployment. In addition, the variance of static positioning and dynamic following positioning is decreased by utilizing the Kalman filtering algorithm. Therefore, the excellent positioning system and its solutions in a pigsty can be provided by our proposed indoor positioning method.
Monitoring fish skin health is essential in aquaculture, where scale loss serves as a critical indicator of fish health and welfare. However, automatic detection of scale loss regions remains challenging due to factors such as uneven underwater illumination, water turbidity, and complex background conditions. To address this issue, we constructed a scale loss dataset comprising approximately 2750 images captured under both clear above-water and complex underwater conditions, featuring over 7200 annotated targets. Various image enhancement techniques were evaluated, and the Clarity method was selected for preprocessing underwater samples to enhance feature representation. Based on the YOLOv8m architecture, we replaced the original FPN + PAN structure with a weighted bidirectional feature pyramid network to improve multi-scale feature fusion. A convolutional block attention module was incorporated into the output layers to highlight scale loss features in both channel and spatial dimensions. Additionally, a two-stage transfer learning strategy was employed, involving pretraining the model on above water data and subsequently fine-tuning it on a limited set of underwater samples to mitigate the effects of domain shift. Experimental results demonstrate that the proposed method achieves a mAP50 of 96.81%, a 5.98 percentage point improvement over the baseline YOLOv8m, with Precision and Recall increased by 10.14% and 8.70%, respectively. This approach reduces false positives and false negatives, showing excellent detection accuracy and robustness in complex underwater environments, offering a practical and effective approach for early fish disease monitoring in aquaculture.
An novel indoor positioning method of pigsty based on Ultra Wideband (UWB) technology is proposed via reasonably arranging the base stations (BTS) and positioning tags, which can provide a strong support for the accurate positioning of specific pig pens in the pig house. The distribution of Dilution of Precision (DOP) values are analysed in different deployment schemes which includes the amount, the height, the deployment shape and distance of BTS, though positioning simulation. Moreover, the positioning effect of dynamic following positioning (DFP), the root-mean-square error (RMSE) and variance of static positioning (SP) are measured. It is noteworthy that Kalman filtering algorithm is adopted to smooth the DFP experimental data for reducing the positioning error. The simulated results show that the DOP values decrease with the increasing of the height and the distance, and these results and laws are consistent with the experimental results. However, the DOP values are reached best when the height is from 1.75 m to 2.75 m, and the distance is over 3.00 m. Furthermore, the DOP values of quadrilateral deployment is smaller than the one of triangle. In addition, variance of SP and DFP are decreased by utilizing Kalman filtering algorithm. Therefore, the excellent positioning system and its solutions in pigsty can be provided by our proposed indoor positioning method.
Reliable 3D visual reconstruction is important for documenting the external quality of navel oranges and supporting future phenotypic analysis. Conventional Structure-from-Motion (SfM) pipelines struggle with the weakly textured characteristics and highly specular surfaces of navel oranges, resulting in incomplete reconstructions, sparse point clouds, and severe visual artifacts. To address these challenges, this study develops a high-fidelity 3D reconstruction framework that integrates deep-learning-based feature matching with 3D Gaussian Splatting (3DGS). For the sparse reconstruction phase, the SuperPoint detector and SuperGlue matcher are employed, leveraging deep feature extraction and graph neural networks to enhance matching robustness on complex surfaces. To address the initialization challenges of 3DGS, a spatial-colorimetric cascade purification strategy—combining pass-through filtering, HSV masking, and statistical filtering—is developed to effectively eliminate noise and provide a high-confidence initialization prior. Subsequently, 3DGS is utilized for dense reconstruction and high-fidelity neural rendering. Experimental results demonstrate that the developed framework significantly outperforms the conventional SfM pipeline in terms of registered image count, point cloud density, and trajectory length. The 3DGS-optimized model achieved improved PSNR, SSIM, and LPIPS values on held-out views, indicating enhanced novel-view rendering quality and visual fidelity. This approach offers an efficient photorealistic digitization solution for weakly textured and highly specular fruits, thereby establishing a reliable visual foundation for future non-destructive phenotypic analysis in smart agriculture.