Stacked AutoEncoder-based Compression of Point Cloud Geometry

Research Square (Research Square), Journal Year: 2025, Volume and Issue: unknown

Published: Jan. 13, 2025

Point clouds have gained widespread application in various fields, but their high resolution often results large data volumes, posing challenges for storage, transmission, and processing. Traditional 2D image or video compression methods are unsuitable due to the spatial irregularity sparseness of point clouds. Inspired by effectiveness autoencoders visual analysis tasks compression, this paper proposes a novel stacked autoencoder-based geometry method By transforming into Morton codes using linear octree structure further encoding them integer sequences, proposed leverages reduce dimensions these achieving both reconstruction quality ratios. Experimental demonstrate that our outperforms many other methods, especially small-size increasing coding depth octree, approach can even achieve lossless results, showcasing its potential as an effective technique

Language: Английский

---

Point Cloud–Based Defect Inspection with Multisensor Fusion and Deep Learning for Advancing Building Construction Quality

Journal of Computing in Civil Engineering, Journal Year: 2025, Volume and Issue: 39(3)

Published: Feb. 10, 2025

Language: Английский

---

Estimating Contaminated Soil Volumes Using a Generative Neural Network: A Hydrocarbon Case in France

Mathematical Geosciences, Journal Year: 2025, Volume and Issue: unknown

Published: May 5, 2025

Language: Английский

---

AI’s Role in Semantic Segmentation for Data-Driven 3D Models of Heritage Structures

Lecture notes in civil engineering, Journal Year: 2025, Volume and Issue: unknown, P. 793 - 808

Published: Jan. 1, 2025

Language: Английский

---

Wheat Head Classification in 3D Point Clouds for Fusarium Head blight Detection

Published: Aug. 30, 2024

Deep learning (DL) has become one of the most efficient tools for data processing in computer vision and is a popular technique tasks such as classification, segmentation, detection. Although these techniques have been applied to with structured grid, 3D point clouds shown proficient results increased popularity due growing availability acquisition devices. This led their application areas robotics, autonomous driving, medicine, agriculture, more. A cloud set points defined metric space, characterized by its unstructured nature. The unstructuredness makes use DL direct challenging object detection an active research topic. important functional method it can simultaneously predict surrounding objects' categories, locations, sizes. In fields like this offers potential analyse various plant attributes, height, biomass, number size relevant organs.Plant recognition represent difficult challenge plants' size, posture, shape, illumination, texture, which vary depending on varieties growth stages. One major presented wheat plants. As fundamental source food, interest analysis increased. Detection spikes help validate spikelet fertility, spike characteristics, evaluate high-yield cultivars. thesis, we created dataset 576 samples multiple plants, manually labeled head classification. Utilizing neural network model specialized clouds, called PointNet, developed identify detect heads. allowed us directly input preserve detailed information. demonstrated test accuracy 80% best model. Finally, CNN-based classification was integrated develop Fusarium Head blight (FHB) fine-tuned disease automatically infected FHB from images spikelets achieved 91% plants set. Extensive cross-validation experiments were performed performance ability promising results. addition, drawbacks proposed analyzed, directions future work are provided.

Language: Английский

---

Energy storage using computer vision: control and optimization of energy storage

Elsevier eBooks, Journal Year: 2024, Volume and Issue: unknown, P. 223 - 239

Published: Oct. 1, 2024

Language: Английский

---