Constraints often exist in the high-dimensional data output in object detection, such as the inverse vector (cos θ, sin θ) of the two-dimensional object and the attitude quaternion of the threedimensional object. The range of each component of the output value of the traditional neural network is unconstrained, which is difficult to meet the needs of practical problems. To solve this problem, this paper designed the transformation network layer according to the high dimensional space transformation theory and constructed a constrained neural network model to detect the pose of objects from a single aerial image. Firstly, in the yolov3 network structure, according to the size of the three field scales, three scale transformation network layers are added correspondingly to implement the constrained unit quaternion field. Secondly, according to the characteristics of quaternion, we proposed a special loss function. Then a new constrained neural network called the quaternion field pose network (qfiled PoseNet) model is constructed, which can predict the probability field of the object and the contained unit quaternion field respectively. Next, The object’s probability field is generated to determine the 2D bounding box of the object, and the unit quaternion field is generated to determine the 3D rotation R. Finally, combining the rotation matrix R and the 2D bounding box of the object to calculate the 3D translation T. We used our method to experiment on DOTA1.5 data set and the HSRC2016 data set respectively. The experimental results show that our method can detect the pose of the object well.

Due to the inherent periodicity of the angle, the direction of the object detected by the current rotating object detection algorithm is fuzzy. In order to solve this problem, this paper proposes a rotating object direction estimation method based on a neural network, which determines the unique direction of the object by predicting the direction vector of the object. Firstly, we use the two components (sin $\theta$, cos $\theta$) of the direction vector and the length and width parameters of the object to express the object model. Secondly, we construct a neural network model to predict the parameters used to express the object model. However, there is a constraint that the sum of the squares of the two components of the direction vector of the object is equal to 1. Because each output element of the neural network is independent, it is difficult to learn the constrained data between such neurons. Therefore, the function transformation model is designed, and the network transformation layer is added. Finally, affine transformation is used to transform the object parameters and carry out regression calculation, so as to detect the object and determine the direction of the object at the same time. This paper uses three sets of data to carry out the experiment, which are DOTA 1.5, HRSC, and UCAS-AOD data sets. It can be seen from the experimental results that for the object with correct ground truth, the proposed method can not only locate the object but also estimate the direction of the object accurately.

The geometric features of ground objects can reflect the shape, contour, length, width, and pixel distribution of ground objects and have important applications in the process of object detection and recognition. However, the geometric features of objects usually present irregular geometric shapes. In order to fit the irregular geometry accurately, this paper proposes the mark clustering point process. Firstly, the random points in the parent process are used to determine the location of the ground object, and the irregular graph constructed by the clustering points in the sub-process is used as the identification to fit the geometry of the ground object. Secondly, assuming that the spectral measurement values of ground objects obey the independent and unified multivalued Gaussian distribution, the spectral measurement model of remote sensing image data is constructed. Then, the geometric extraction model of the ground object is constructed under the framework of Bayesian theory and combined with the reversible jump Markov chain Monte Carlo (RJMCMC) algorithm to simulate the posterior distribution and estimate the parameters. Finally, the optimal object extraction model is solved according to the maximum a posteriori (MAP) probability criterion. This paper experiments on color remote sensing images. The experimental results show that the proposed method can not only determine the position of the object but also fit the geometric features of the object accurately.

Recent years have seen significant advancements in computer vision, spearheaded by the advent of deep learning networks. However, the rigid architectures of these networks limit their potential. This study investigates how altering existing networks architecture can enhance performance. However, modifying layers in a model leads to a mismatch in pre-trained weights, resulting in the loss of a considerable amount of information embedded in the pre-trained model. Concurrently, an inefficient post-training process can lead to a significant waste of time if it does not fully leverage the pre-trained model weights. We propose a novel Homomorphic Model Transformation (HMT), based on functional algebraic identity, to overcome this challenge. HMT ensures that the model’s output remains identical to that of the original model even after expanding its architecture. This enables post-training enhancements to commence from the highest point of the original model. Experiment results reveal that HMT significantly improve training efficiency, reducing post-training time by over 90%. Additionally, HMT significantly enhances model performance metrics in a short timeframe, demonstrating a 4.70% increase in AP on the AI-TOD dataset, a 6.88% improvement on the DOTA1.5 dataset, a 3.05% enhancement on the COCO2017 dataset, and a 4.89% elevation on the MRSAText dataset.

Oriented object detection in remote-sensing imagery benefits from a geometric representation that remains well-defined across localization, label assignment, and post-processing. We present Ellipse-ABC, a YOLOv8-based detector that directly predicts the coefficients of a positive-definite quadratic form while retaining the original branch for center estimation. The resulting matrix supplies Gaussian-compatible localization supervision and is analytically decoded into standardized OBB geometry for task-aligned assignment, rotated non-maximum suppression, and benchmark reporting, thereby reducing representation changes across the pipeline. Primary accuracy is evaluated with polygon area IoU between predicted and ground-truth-oriented bounding boxes. After rerunning the native baseline under the same dataset-specific training and augmentation recipes used by the ABC configurations, the complete ABC + GWD configuration changes polygon-IoU mAP50/mAP50-95 by +2.16/−1.17 points on UCAS-AOD and −1.23/+0.50 points on HRSC2016 across seeds 0, 1, and 2. An ABC + ProbIoU loss control performs better than ABC + GWD on HRSC2016 but worse on UCAS-AOD, indicating that the geometry-loss choice is dataset-dependent. The representation adds 198 parameters and no more than 0.0018 GFLOPs at the evaluated input sizes. Five repeated batch-1 measurements nevertheless show a 6.6–8.8% end-to-end latency increase and higher peak memory on DOTA-v1.5. These results support Ellipse-ABC as a geometry-consistent representation with small model-level overhead, without implying a universal accuracy gain.

Flood assessments require reproducible links between spatial products and the measurements ultimately reported to analysts, yet a large language model (LLM) agent may complete a geographic information system (GIS) workflow without generating the required product or traceable claim. We present GeoFloodAgent, an LLM-orchestrated flood-GIS workflow that combines typed planning and deterministic GIS tools with a separate evaluation of runtime disposition and post-run alignment with task-specific product, lineage, and action contracts. We evaluated 80 Chinese-language tasks from the Zhengzhou (2021) and Forlì (2023) flood cases under seven configurations and three repetitions, yielding a registry-remediated composite of 1680 assignments. In the Full configuration, 212 assignments (88.3%) were accepted and contract-aligned, whereas 16 accepted assignments did not align with their registered contracts; all 16 concerned clarification or refusal actions. Three runs were rejected despite retained contract-aligned evidence. The cluster-weighted Full-Tool Agent differences were −1.15 percentage points for accepted contract mismatch and +1.07 percentage points for contract-aligned acceptance, with empirical intervals spanning benefit and harm. Under frozen Zhengzhou inputs, deterministic outputs were sensitive to the minimum SAR object-size setting: relative to 80 pixels, candidate flood extent changed by +105.6% at 40 pixels and −30.9% at 120 pixels, whereas a 20 m simplification tolerance reproducibly failed during geometric union. These configuration-specific results do not validate hydrological accuracy or identify universal optimal defaults. These findings show that completion rate alone is inadequate for evaluating LLM-orchestrated flood GIS: runtime disposition and evidence alignment should be reported jointly. The study evaluates a finite, study-authored benchmark and does not establish hydrological accuracy, operational decision validity, or open-world error detection.