Tree height is a crucial structural parameter in forest inventory as it provides a basis for evaluating stock volume and growth status. In recent years, close-range photogrammetry based on smartphone has attracted attention from researchers due to its low cost and non-destructive characteristics. However, such methods have specific requirements for camera angle and distance during shooting, and pre-shooting operations such as camera calibration and placement of calibration boards are necessary, which could be inconvenient to operate in complex natural environments. We propose a tree height measurement method based on three-dimensional (3D) reconstruction. Firstly, an absolute depth map was obtained by combining ARCore and MidasNet. Secondly, Attention-UNet was improved by adding depth maps as network input to obtain tree mask. Thirdly, the color image and depth map were fused to obtain the 3D point cloud of the scene. Then, the tree point cloud was extracted using the tree mask. Finally, the tree height was measured by extracting the axis-aligned bounding box of the tree point cloud. We built the method into an Android app, demonstrating its efficiency and automation. Our approach achieves an average relative error of 3.20% within a shooting distance range of 2–17 m, meeting the accuracy requirements of forest survey.
In the context of expensive constraint multi-objective problems, it is evident that the feasible domain shapes and sizes of different problems vary considerably. The difficulty in finding optimal solutions presents a significant challenge in ensuring the surrogate-assisted evolutionary algorithm’s feasibility, convergence, and diversity. To more effectively address the distinctive characteristics of the feasible domain and objective function across a range of problems, we have developed a Kriging-based surrogate-assisted evolutionary algorithm tailored to the current population’s preferences. The algorithm can optimize the population according to the current population’s requirements. Additionally, considering the varying degrees of accuracy observed in the surrogate models at different stages, this paper employs a dynamic approach to the number of surrogate model evaluations, contingent on the accuracy of the current surrogate model. Two types of Pareto frontier search are distinguished: unconstrained and constrained. Moreover, distinct fill sampling strategies are devised in accordance with the specific optimization requirements of the current population. After assessing the proposed solutions, the discrepancy between the actual fitness value and the surrogate model’s prediction is calculated.The discrepancy is used to modify the number of evaluations conducted on the surrogate model. In order to illustrate the algorithm’s efficacy, it is benchmarked against the current state-of-the-art algorithms on various test problems. The experimental results demonstrate that the proposed algorithm performs better than other advanced methods.
Time series forecasting is extensively utilised in meteorology, transportation, finance, and industrial domains. Precisely recognising cyclical trends and abrupt local changes in time series is essential for enhancing forecasting accuracy. Frequency-domain representations are adept at identifying periodic traits, whereas time-domain approaches are superior for spotting localised quick changes. Traditional techniques sometimes prioritise a single domain, overlooking the advantages of integration. This paper introduces a novel hybrid model, FFTNet, that simultaneously pulls characteristics from both domains to optimise their respective benefits. Theoretical examination indicates that the 2D CNN utilises dual-axis convolution kernels to jointly describe global cross-patch structures and local temporal patterns, whilst the frequency-domain MLP enhances spectral components in accordance with Parseval’s Theorem and the Convolution Theorem. A frequency-domain MLP is employed to discern periodic and trend characteristics, while a 2D CNN in the time domain identifies localised abrupt changes. This hybrid methodology differentiates itself from previous methods that depend exclusively on a single domain, providing a more thorough comprehension of the underlying patterns in time series data. Experiments on seven real-world datasets indicate that FFTNet surpasses existing techniques, attaining state-of-the-art performance with enhancements of 11.8% and 4.7% in MSE and MAE, respectively.
Multimodal multiobjective optimization problems (MMOPs) are widely encountered in real-world applications. While numerous evolutionary algorithms have been developed to locate equivalent Pareto-optimal solutions, existing Multimodal Multiobjective Evolutionary Algorithms (MMOEAs) often struggle to handle large-scale decision variables and sparse Pareto sets due to the curse of dimensionality and unknown sparsity. To address these challenges, this paper proposes a novel approach named MASR-MMEA, which stands for Large-scale Sparse Multimodal Multiobjective Optimization via Multi-stage Search and Reinforcement Learning (RL)-assisted Environmental Selection. Specifically, to enhance search efficiency, a multi-stage framework is established incorporating three key innovations. First, a dual-strategy genetic operator based on improved hybrid encoding is designed, employing sparse-sensing dynamic redistribution for binary vectors and a sparse fuzzy decision framework for real vectors. Second, an affinity-based elite strategy utilizing Mahalanobis distance is introduced to pair real vectors with compatible binary vectors, increasing the probability of generating superior offspring. Finally, an adaptive sparse environmental selection strategy assisted by Multilayer Perceptron (MLP) reinforcement learning is developed. By utilizing the MLP-generated Guiding Vector (GDV) to direct the evolutionary search toward efficient regions and employing an iteration-based adaptive mechanism to regulate genetic operators, this strategy accelerates convergence. Furthermore, it dynamically quantifies population-level sparsity and adjusts selection pressure through a modified crowding distance mechanism to filter structural redundancy, thereby effectively balancing convergence and multimodal diversity. Comparative studies against six state-of-the-art methods demonstrate that MASR-MMEA significantly outperforms existing approaches in terms of both solution quality and convergence speed on large-scale sparse MMOPs.
Three-dimensional mesh pose transfer transforms the pose of a source model into the pose of a reference model while preserving the source model’s identity (body detail). It has tremendous potential in computer graphics tasks. Current neural network-based methods primarily focus on extracting pose and body features, not entirely using the articulated body structure of humans and animals. We propose an end-to-end pose transfer network based on skinning deformation to address these issues. This network first extracts skinning weights and model joint features. Then, they are decoded to transfer the source model to a pose similar to the reference model while preserving the features of the source model. During feature extraction, we utilize the features of the k-nearest neighborhoods and one-ring neighborhoods to enable the network to learn the body details of the model better. Additionally, we apply skinning weights and joint features to capture the variation in the source model pose compared to the reference model pose and then use a decoding network to obtain the target model, replacing linear blend skinning. We conducted experiments on datasets such as SMPL, SMAL, FAUST, DYNA, and the MG dataset to provide empirical evidence and demonstrate that our method is the best in quantitative experiments. Our method efficiently transfers poses while better preserving the identity of the source model.