Handheld ophthalmoscopes often produce low-quality fundus images, which negatively affect the performance of deep learning models in diagnosing fundus diseases. Existing methods for fundus image enhancement mainly focus on visual improvement rather than optimizing the enhancement outcomes for specific diagnostic tasks.We propose a method for processing low-quality fundus images and diagnosing fundus diseases using an enhancement model. To address the lack of paired low/high-quality glaucoma images, we artificially degraded images from the BrG public dataset. We then trained the RFormer enhancement model using transfer learning to transfer knowledge from images captured by traditional medical equipment to those captured by portable devices.The proposed method is designed with downstream diagnostic tasks in mind, creating a feedback loop that uses diagnostic model loss to guide real-time updates to the enhancement model. This approach not only boosts diagnostic performance but also enhances the perceived quality of the output images.
As Whole Slide Images(WSI) are characterized by their extremely high resolution and lack of pixel-level annotation, Multiple Instance Learning(MIL) has become an effective tool for addressing WSI classification challenge. However, current MIL methods typically rely on the independent and identically distributed assumption, neglecting correlation among different instances and only analyze WSI at a single scale, failing to effectively utilize the complementarity between multi-scale features. To enhance classification performance and obtain more accurate diagnostic results, we propose a MIL model based on multi-scale Graph-Transformer framework. It utilizes graph convolutional network to capture spatial, semantic, and scale correlations among patches, and performs graph pooling to select discriminative patches in order to reduce computational costs. Additionally, we introduce a multi-scale feature fusion module aimed at integrating features across different scales. The experimental results on the TCGA-NSCLC and TCGA-RCC datasets show that our method achieves consistent high performance for different cancer types, obtaining accuracy of 0.9281 and 0.9508 on the above two datasets, respectively. Compared with other comparison methods, our method has surperior classification performance.
Vision based runway identification using standard ‘marked runway’ images captured from a fixed wing unmanned aerial vehicle (UAV) through onboard monocular or stereovision sensor is presented in this paper. An innovative convolutional neural netwok (CNN) based ‘You Only Look Once’ (YOLO) object detection algorithm has been exploited to detect the runway threshold region during approach segment of landing of a UAV. This threshold region is defined as Region of Interest (ROI). ORB-SLAM3 algorithm with visual odometry has been configured for feature detection of runway and feature correspondence between successive camera frame’s ‘feature points’ to compute rotational and translational matrices and compute ‘position and orientation’ (POSE) of camera in ORB-SLAM coordinate frame. Perspective n point (PnP) algorithm has been configured for camera pose estimation in runway coordinate frame. Feature detector based image processing algorithm with monocular (with scaling constraint) and stereo vision constraint is used to establish the relation between 3D corner points of ROI and corresponding detected corner feature points. Based on the detected runway, a runway coordinate frame is constructed at ROI center and UAV onboard camera pose in runway coordinate frame is computed in real time.
In this paper, we present a risk-aware motion planning framework for resilient navigation in occupancy grid maps. Navigating in unknown environments involves complex interactions with the surroundings that must be effectively managed by the robot. Recently, these interactions are frequently expressed as risk constraints, where risk is defined as potential threats that could hinder the robot from accomplishing its objectives. However, when the risk constraint involves nonlinearities, common numerical solvers are severely hampered in finding a feasible solution and are prone to failure. Therefore, we present a novel risk-aware navigation strategy based on motion primitives and the Nonlinear Model Predictive Control (NMPC) method to address nonlinear risk constraints within discrete maps. We demonstrate the effectiveness of our approach through a practical application of a robust risk assessment method that takes into account both the state of the environment and the state of the robot. In addition to enhancing the decision-making capabilities of the robot, our framework offers a more resilient motion planning process that enables the robot to navigate risky scenarios where standard optimizers are likely to fail and lead to dangerous trajectories.
Understanding the environment is crucial for the autonomous navigation of vehicles. Accurately identifying and removing dynamic objects that cause occlusions and noisy pose issues is crucial to the task. The casualty collection point (CCP) is a designated location for treating casualties during disasters. These sites are commonly located in open fields to ensure that the injured receive timely and appropriate care. Therefore, the construction of a 3D map in CCP may encounter additional challenges. In this paper, we introduces a novel algorithm that integrates a learning-based multi-object detection network with a Kalman Filter tracking framework to remove dynamic object traces from the 3D map building, particularly in CCP scenarios, and it redirects attention on the velocity attribute of dynamic objects at the object-level scale. Additionally, we contribute a new dataset specifically designed for the CCP scenario. Comparative experiments conducted on the SemanticKITTI dataset and CCP dataset show that our proposed method achieves the state-of-the-art performance in removing traces of dynamic objects on 3D Map. Dataset is available at https://github.com/haidongwang96/ccp_dataset.
