Quantitative remote sensing estimation is critical for environmental monitoring, providing continuous measures of vegetation indices, canopy height, and carbon stock. Traditional radiative-transfer models and empirical regressions require expert knowledge and generalize poorly, while deep learning methods remain task-specific. We propose SatelliteCalculator+, a DINOv3-powered multi-task foundation model for continuous regression of spectral and structural variables. The framework combines prompt-driven cross-attentive adapters with lightweight MLP decoders, enabling efficient dense prediction from frozen features. To overcome limited supervision, we synthesize over one million paired samples from SPOT 6/7 imagery using physically defined formulas. On the Open-Canopy dataset, SatelliteCalculator+ achieves competitive accuracy across eight ecological variables while reducing inference cost, demonstrating the promise of self-supervised transformers and scalable multi-task learning for large-scale Earth observation.

Transformer-based detectors have advanced small-object detection, but they often remain inefficient and vulnerable to background-induced query noise, which motivates deep decoders to refine low-quality queries. We present HELP (Heatmap-guided Embedding Learning Paradigm), a noise-aware positional-semantic fusion framework that studies where to embed positional information by selectively preserving positional encodings in foreground-salient regions while suppressing background clutter. Within HELP, we introduce Heatmap-guided Positional Embedding (HPE) as the core embedding mechanism and visualize it with a heatbar for interpretable diagnosis and fine-tuning. HPE is integrated into both the encoder and decoder: it guides noise-suppressed feature encoding by injecting heatmap-aware positional encoding, and it enables high-quality query retrieval by filtering background-dominant embeddings via a gradient-based mask filter before decoding. To address feature sparsity in complex small targets, we integrate Linear-Snake Convolution to enrich retrieval-relevant representations. The gradient-based heatmap supervision is used during training only, incurring no additional gradient computation at inference. As a result, our design reduces decoder layers from eight to three and achieves a 59.4% parameter reduction (66.3M vs. 163M) while maintaining consistent accuracy gains under a reduced compute budget across benchmarks. Code Repository: https://github.com/yidimopozhibai/Noise-Suppressed-Query-Retrieval.

Generative AI presents transformative potential across various domains, from creative arts to scientific visualization. However, the utility of AI-generated imagery is often compromised by visual flaws, including anatomical inaccuracies, improper object placements, and misplaced textual elements. These imperfections pose significant challenges for practical applications. To overcome these limitations, we introduce Yuan, a novel framework that autonomously corrects visual imperfections in text-to-image synthesis. Yuan uniquely conditions on both the textual prompt and the segmented image, generating precise masks that identify areas in need of refinement without requiring manual intervention—a common constraint in previous methodologies. Following the automated masking process, an advanced inpainting module seamlessly integrates contextually coherent content into the identified regions, preserving the integrity and fidelity of the original image and associated text prompts. Through extensive experimentation on publicly available datasets such as ImageNet100 and Stanford Dogs, along with a custom-generated dataset, Yuan demonstrated superior performance in eliminating visual imperfections. Our approach consistently achieved higher scores in quantitative metrics, including NIQE, BRISQUE, and PI, alongside favorable qualitative evaluations. These results underscore Yuan's potential to significantly enhance the quality and applicability of AI-generated images across diverse fields.

Adversarial learning has achieved great success for unsupervised domain adaptation (UDA). Existing adversarial UDA methods leverage the predicted discriminative information with Nuclear-norm Wasserstein discrepancy for feature alignment. However, the limited memory space makes it very difficult to accurately calculate the Nuclear-norm, which hinders domain adaptation. To address this challenge, we propose a multi-batch Nuclear-norm adversarial network, termed as MBAN. Specifically, we build a dynamic queue to cache features, which encourages to generate a large and consistent output matrix, enabling accurate calculation of the Nuclear-norm. Then, the multi-batch Nuclear-norm discrepancy is proposed, which can effectively improve the transferability and discriminability of the learned features. Experimental results show that MBAN could achieve significant performance improvement, especially when the number of categories is quite large. Code is available at https://github.com/peiwang0518/Multi-BAN.

Adversarial domain adaptation has achieved impressive performances for unsupervised domain adaptation (UDA). However, existing adversarial UDA methods often rely on multiple domain discriminators to capture diverse patterns, which limit their scalability and resulting in dispersed features. To address these issues, we propose a novel method called Class-aware Prototypical Adversarial Network (CaPAN), which efficiently extracts transferable and discriminative features. Specifically, our class-aware adversarial learning employs a single multi-class discriminator to capture various patterns, aligning class-level features. Furthermore, to enhance the discriminative ability of our model, we introduce a prototypical domain discriminator to enhance the discriminatively of the learned features by aligning target sample towards prototypes (centers of each class), resulting in a more compact feature space. Extensive experiments validate the effectiveness of CaPAN, which can also be integrated as a regularization technique for existing methods to further improve their performance. Code is available at https://github.com/YuZhenyuLindy/CaPAN.

Edge detection is a critical technology in processing potential field data, enabling the rapid identification of geological body edges using magnetic anomaly data. Traditional methods for detecting edges in magnetic data are known for their simplicity and efficiency; however, they suffer from low resolution, poor robustness, and a lack of depth information. In recent years, the application of deep learning (DL) to edge detection in field data has enhanced both the resolution and robustness of these methods. Nonetheless, these approaches still fail to determine the buried depths of geological body edges. To address this issue, this study has developed an edge detection method called multiconstraint DL based on UNet++, which not only identifies the edge positions but also ascertains the buried depths of geological bodies. The study proposes the multiconstraint loss function for edge detection and employs the Dice loss function and the mean square error (mse) loss function to jointly supervise the network’s training. Subsequently, the label design was revised to include depth information on the geological body, enabling the DL method to accurately determine the buried depths of geological bodies. Analysis of the test model’s detection results reveals that the edge detection method based on UNet++ can precisely identify both the edge positions and the exact buried depths of geological bodies, which makes up for the shortcoming that the traditional edge detection method does not contain the depth information. Moreover, this method resolves the issue of discontinuity found in traditional DL edge detection. Finally, the method was applied to actual aeromagnetic data from the Dandong area in Liaoning Province, successfully identifying the edge positions of the main mining area and providing targeted regions for further exploration.

With the continuous growth of remote sensing data and the application of fast and effective atmosphere remote sensing inversion algorithm, this paper proposes a PM10 fast inversion approach based on Spark platform which uses Apache Spark as the analytics engine and integrates with the traditional atmospheric remote sensing inversion algorithm. We first store aerosol data which is MYD04_3K from NASA into HDFS. Then the inversion algorithm is combined with Spark via the function interface to realise rapid atmospheric remote sensing inversion. The experimental results based on Spark platform are compared with those obtained from the traditional physical hardware. The results prove that the proposed atmospheric remote sensing inversion method based on Spark has high efficiency.

Robust path following capability is mandatory for an unmanned boat to perform tasks autonomously. This paper presents a mixed H∞/H2based control which enables the boat to follow paths with connected straight line segments and circular arcs in the absence of yaw angle measurement. The performance of the strategy is evaluated in field test and will be discussed in the paper.

In this paper we try to develop a host-based system and study actual sea trials via rudder based roll control method. To authors' best knowledge, the boat we investigated is the smallest among those reported in the literature. An autonomous boat model is obtained by a system identification approach. The identified system is designed with frequency-shaped sliding mode control. The control scheme is composed of a sliding mode observer and a sliding mode controller. The stability and reachability of the switching function are proved by Lyapunov theory. Computer simulations and experiment show that successful course keeping and roll reduction results are achieved.

In this paper, we describe a miniature flight platform weighing less than 700 grams and capable of way-point navigation, trajectory tracking, precise hovering and automatic takeoff and landing. In an effort to make advanced autonomous behaviors available to mini and micro rotorcraft, a lightweight/portable and inexpensive Guidance, Navigation, and Control system (GN&C) was developed. To compensate for the weaknesses of the low-cost equipment, we put our efforts in obtaining a reliable model-based nonlinear controller. The GN&C system was implemented on a small four rotor helicopter which has undergone an extensive program of flight tests, resulting in various flight behaviors under autonomous control from takeoff to landing. Flight test results that demonstrate the operation of the GN&C algorithms on a real MAV are presented.

This paper presents the development of vision based autonomous helicopter at Chiba University. The platform is designed to enable autonomous flight of a small-scale helicopter in unstructured environment. 3D vision system is chosen for the perception and measurement purpose. The whole system is designed with the consideration of payload, hardware availability, and software performance. The hardware is built with COTS (commercially off the shelf) products. The core control software is built on RTLinux for its real time performance. Based on the developed platform, we have studied the application of vision system for landing control. The experimental results are shown in the paper

In this paper we introduce a new approach to visually manipulate, with the aid of a robot manipulator, a part placed randomly on a rotating turntable. The highlight of our approach is that the camera and the robot end effector are both assumed to be uncalibrated. We only assume that the height of the robot end effector is known. Our approach utilizes virtual rotation of the camera via image processing not previously introduced in the literature. Finally, the tracking scheme is implemented by planning the error and gradually forcing it to zero while maintaining the torque controls within acceptable limits. This way we demonstrate a new visually guided analytical tracking scheme.

Discusses the problem of using visual and other sensors in the manipulation of a part by a robotic manipulator in a manufacturing workcell. The authors' emphasis is on the part localization problem involved. The authors introduce a new sensor-fusion approach which fuses sensory information from different sensors at various spatial and temporal scales. Relative spatial information obtained from processing of visual information is mapped to absolute taskspace of the robot through fusing of information from an encoder. Data obtained this way can be superimposed upon data obtained from displacement based vision algorithms at coarser time scales to improve overall reliability. Tracking plans reflecting sensor fusion are proposed. The localization of a part by spatial sensor fusion is experimentally demonstrated to be able to give required fast and accurate part localization.

A multi-sensor based planning and control scheme for robotic manufacturing is presented in this paper. The proposed approach fuses sensory information from various sensors at different temporal and spatial scales in an event-based planning and control scheme. By combining the measurement of an encoder sensor, relative spatial information obtained from processing of visual measurement is mapped to the absolute task-space of the robot, delayed data obtained from a displacement based vision algorithm that represent absolute part position measurement is brought to up to date. A four-step approach to planning and control of a robotic manipulator is discussed. An event-driven tracking and control scheme that is based on multi-sensor information is given. The approach is illustrated by considering a manufacturing workcell where the manipulator is commanded to pick up a part on a disc conveyor under the guidance of computer vision.