Background: Accurate retinal layer segmentation in optical coherence tomography (OCT) images is crucial for quantitatively analyzing age-related macular degeneration (AMD) and monitoring its progression. However, prev...
详细信息
Background: Accurate retinal layer segmentation in optical coherence tomography (OCT) images is crucial for quantitatively analyzing age-related macular degeneration (AMD) and monitoring its progression. However, previous retinal segmentation models depend on experienced experts and manually annotating retinal layers is time-consuming. On the other hand, accuracy of AMD diagnosis is directly related to the segmentation model's performance. To address these issues, we aimed to improve AMD detection using optimized retinal layer seg-mentation and deep ensemble ***: We integrated a graph-cut algorithm with a cubic spline to automatically annotate 11 retinal boundaries. The refined images were fed into a deep ensemble mechanism that combined a Bagged Tree and end-to-end deep learning classifiers. We tested the developed deep ensemble model on internal and external datasets. Results: The total error rates for our segmentation model using the boundary refinement approach was signifi-cantly lower than OCT Explorer segmentations (1.7% vs. 7.8%, p-value = 0.03). We utilized the refinement approach to quantify 169 imaging features using Zeiss SD-OCT volume scans. The presence of drusen and thickness of total retina, neurosensory retina, and ellipsoid zone to inner-outer segment (EZ-ISOS) thickness had higher contributions to AMD classification compared to other features. The developed ensemble learning model obtained a higher diagnostic accuracy in a shorter time compared with two human graders. The area under the curve (AUC) for normal vs. early AMD was 99.4%.Conclusion: Testing results showed that the developed framework is repeatable and effective as a potentially valuable tool in retinal imaging research.
ABSTRACTABSTRACTAerial imagery is important in remote sensing applications. Unmanned aerial vehicle (UAV) has a wide range of applications in remote sensing and presents a substantial cost-effective solution when moni...
详细信息
ABSTRACTABSTRACTAerial imagery is important in remote sensing applications. Unmanned aerial vehicle (UAV) has a wide range of applications in remote sensing and presents a substantial cost-effective solution when monitoring objects on the earth’s surface. Moreover, object detection and classification are important aspects of global information system, especially for remote sensing applications and power line monitoring, which are essential for the proper distribution of electricity to consumers. Manual inspection consumes much time and involves risk, especially in remote areas that host dangerous wildlife; hence, UAV-based approaches are more feasible for such monitoring. The authors propose an UAV approach that utilises a digital surface model and incorporates a stereo matching algorithm based on UAV stereo images. The proposed algorithm was based on a graph-cut (GC) algorithm that measured the disparity map. Results were compared with well-known algorithms; including, for example, global and local stereo matching algorithms. The proposed solution introduces and integrates ordering constraints along with a submodular energy minimisation function to/with the GC algorithm to enhance performance. The authors measured sensitivity and recall for all parameters against ground truth data for differently cropped images of 16 power poles. Results showed that the proposed model performed more accurately compared to extant methods.
We proposed a novel deep convolutional neural network based species recognition algorithm for wild animal classification on very challenging camera-trap imagery data. The imagery data were captured with motion trigger...
详细信息
ISBN:
(纸本)9781479957521
We proposed a novel deep convolutional neural network based species recognition algorithm for wild animal classification on very challenging camera-trap imagery data. The imagery data were captured with motion triggered camera trap and were segmented automatically using the state of the art graph-cut algorithm. The moving foreground is selected as the region of interests and is fed to the proposed species recognition algorithm. For the comparison purpose, we use the traditional bag of visual words model as the baseline species recognition algorithm. It is clear that the proposed deep convolutional neural network based species recognition achieves superior performance. To our best knowledge, this is the first attempt to the fully automatic computer vision based species recognition on the real camera-trap images. We also collected and annotated a standard camera-trap dataset of 20 species common in North America, which contains 14,346 training images and 9, 530 testing images, and is available to public for evaluation and benchmark purpose.
Late blight and early blight are the most destructive diseases for potatoes. It is valuable to distinguish diseases and their degrees of infection on potato leaves for timely prevention. This study investigated an acc...
详细信息
Late blight and early blight are the most destructive diseases for potatoes. It is valuable to distinguish diseases and their degrees of infection on potato leaves for timely prevention. This study investigated an accurate recognition method for detecting the disease type and degree of infection from potato leaf images. To segment the leaf from the images efficiently and accurately, an automatic scheme for the graph-cut algorithm is developed. The seeds of the foreground were extracted by Otsu thresholding, and the seeds of the background were extracted by color statistical thresholding on a* and b* components. To remove the backgrounds that have similar color as the infected patch, the superpixels that neighbor the outline of the leaf will be iteratively eliminated when their entropies are far from those of the major part of the leaf. Then, the color features were extracted from the individual channels of the L*a*b* on the refined region of interest (ROI), and the texture features were extracted using a local binary pattern (LBP). Finally, four classifiers based on the k-nearest neighbor (k-NN), support vector machine (SVM), artificial neural network (ANN) and random forest (RF) methods were adopted to evaluate the performance for recognition of potato disease. The performance of the proposed method was evaluated on 2840 images of healthy and diseased potato leaves. The segmentation results showed that the average intersection over union (IoU) was 93.70% for the five classes. For disease classification, the SVM classifier achieved the highest overall accuracy of 97.4% compared with k-NN, ANN and RF. For the degree of infection classification, and the SVM classifier achieved the highest overall accuracy of 91.0%. To enhance the classification performance, a combination of six types of features was evaluated. The results showed that SVM achieved the highest overall accuracy of 92.1% with the combinations of a local binary pattern (LBP) on the a* component, LBP on the
暂无评论