Search

Recently, atrous spatial pyramid module and encoder-decoder structure are widely investigated for many computer vision tasks to address the scale variation challenge in deep convolutional networks. The atrous spatial pyramid module aims at capturing local and global cues together with various sampling rates in convolutional or pooling layers. At the same time, encoder-decoder structure propagates context features from low-resolution, strong-semantic features to high-resolution, weak-semantic ones, while maintaining the detail object boundaries. However, the aforementioned two strategies have their own drawbacks, and the previous object detectors only employ one of them to handle scale variation. In this work, we propose to cooperate atrous spatial pyramid convolution (ASPC) with encoder-decoder structure (ED) for object detection, termed ASPC-ED, which combines the complementary advantages from both modules in an end-to-endfashion. Specifically, the proposed method is consist of three components: encoder, ASPC and decoder. The extensive experiments on PASCAL VOC and MS COCO benchmarks demonstrate that our method with various backbone achieves the state-of-the-art results for object detection, instance segmentation and panoptic segmentation.

In order to decrease the vulnerability of ground support aircraft under fragment/projectile attack, a vulnerability model of a ground support aircraft was established, and the vulnerability of the ground support aircraft in the typical attack direction was calculated. According to the geometric shape of the ground support aircraft and the distribution of critical components, the ground support aircraft is divided into different areas, and the armor protection efficiency of each area is assessed. The results show that the cockpit area, the front fuselage fuel tank area, and the rear fuselage fuel tank area have high protection efficiency, and these areas can be protected first when the armor weight is limited. On this basis, flexible use of vulnerability reduction measures, targeted reduction of highly vulnerable components, and repeated iterative analysis to achieve vulnerability index requirements are performed. Finally, a numerical simulation test was carried out on the A-10 ground support aircraft. After the vulnerability was reduced, the index design requirements were met under the B and C level kills, which verified the feasibility of the proposed method.

In intelligent manufacturing system, a systematic method of constructing stereo vision using neural network is proposed, and the operation and control mechanism in practical operation is put forward. In particular, the concept of calculation process is established. And this method overcomes the complexity of a large number of calculations. And it has significant advantages in the system scientifically, the operation speed and the operability.

The continuous zoom multi configuration method is developed, and the correlative model is established, which overcomes the limitations of traditional methods, and realizes the expected system evaluation criteria，such as MTF, REA, RMS, etc..through the balance between value function, optimization algorithm and model parameterization.

Multidisciplinary design optimization (MDO), as an advanced and effective theory and method to solve the design problems of many complex products and systems, has attracted more and more attention. Based on the analysis of product design activities, the effective theory and engineering practice of complex systems are discussed.

Shrinking optical components and systems without compromising performance has been a challenge. New sensors and components are developing the methodology in research advanced multiple configuration systems. The effect of system evaluated criteria that Modulation Transfer Function (Modulation Transfer Function, the MTF), Energy concentration (Radial Energy Analysis, REA), diffusion (Spot Diagram, RMS) can support our issue qualitatively. Results showed that can not only change systems weight, size and structural stability, but also increase freedom in these kinds design.

Hyperspectral anomaly detection (HAD) is a research endeavor of high practical relevance within remote sensing scene interpretation. In this work, we propose an unsupervised approach, dual feature extraction network (DFEN) for HAD, to gradually build up ever-greater discrimination between the original data and background. In particular, we impose an end-to-end discriminative learning loss on two networks. Among them, adversarial learning aims to keep the original spectrum while Gaussian constrained learning intends to learn the background distribution in the potential space. To extract the anomaly, we calculate spatial and spectral anomaly scores based on mean squared error (MSE) spatial distance and orthogonal projection divergence (OPD) spectral distance between two latent feature matrices. Finally, the comprehensive detection result is obtained by a simple dot product between two domains to further reduce the false alarm rate. Experiments have been conducted on eight real hyperspectral data sets captured by different sensors over different scenes, which show that the proposed DFEN method is superior to other compared methods in detection accuracy or false alarm rate.