As the typical man-made objects, buildings play a special role in remote sensing images. This paper proposes a building scene recognition method based on deep multiple instance convolutional neural network. Scene clas...
详细信息
ISBN:
(纸本)9781450371483
As the typical man-made objects, buildings play a special role in remote sensing images. This paper proposes a building scene recognition method based on deep multiple instance convolutional neural network. Scene classification is formulated as a multiple instance learning (MIL) problem in which local scene regions are regarded as instances and assigned with different labels. Through a trainable MIL pooling based on a spatial attention mechanism to select the most relevant instances adaptively and produce the scene-level predictions. Experimental results on UCM data set and smalltargetsdata set collected based on the GF-2 show that the method can improve the accuracy of building scene recognition using high resolution remote sensing image. Particularly, it solves the problem that it is difficult to detect the small buildings correctly and provide a new idea of scene recognition of buildings.
The three-dimensional high-resolution imaging sonar uses multi-beam technology combined with synthetic aperture technology to achieve centimeter-level imaging of small underwater targets. It has a good imaging effect ...
详细信息
As one of the source sensors of maritime intelligent traffic network, radar plays an important role in maritime monitoring and early warning. The number of features extracted by the traditional maritime radar target d...
详细信息
As one of the source sensors of maritime intelligent traffic network, radar plays an important role in maritime monitoring and early warning. The number of features extracted by the traditional maritime radar target detection method in feature domain is small, and the sea clutter and target echo features are often not linearly separable, so the radar detection performance is seriously affected by sea clutter. To solve this problem, this paper combines the time-frequency domain processing method with the residual neural network, and uses the time-frequency transform method to improve the signal-to-clutter ratio (SCR) and the degree of difference between sea clutter and target echo. On this basis, the high-dimensional time-frequency spectrum features are extracted by using the residual neural network to improve the utilization rate of radar echo information, form a high-dimensional feature space of sea clutter and target echo that are non-linearly separable, and realize the binary classification of sea clutter and target echo. It is verified by the measured data of X-band radar that the proposed target detection method can extract the deep features of the time-frequency spectrum of radar echo, has a high classification accuracy even in the case of low SCR, and has the potential to detect weak targets in the background of strong sea clutter. In addition, the influence of different time-frequency transform methods and polarization modes on target detection performance is further analyzed. The comparative study shows that different time-frequency transform methods and polarization modes have little influence on the classification accuracy of the proposed method. In comparison, under the conditions of fractional Fourier transform and cross-polarization, the proposed method has higher classification accuracy.
Many approaches in CEM rely on the decomposition of complex radiation and scattering behavior with a set of basis vectors. Accurate estimation of the quantities of interest can be synthesized through a weighted sum of...
详细信息
Many approaches in CEM rely on the decomposition of complex radiation and scattering behavior with a set of basis vectors. Accurate estimation of the quantities of interest can be synthesized through a weighted sum of these vectors. In addition to basis decompositions, sparse signalprocessing techniques developed in the CS community can be leveraged when only a small subset of the basis vectors are required to sufficiently represent the quantity of interest. We investigate several concepts in which novel bases are applied to common electromagnetic problems and leverage the sparisty property to improve performance and/or reduce computational burden. The first concept explores the use of multiple types of scattering primitives to reconstruct scattering patterns of electrically large targets. Using a combination of isotropic point scatterers and wedge diffraction primitives as our bases, a 40 reduction in reconstruction error can be achieved. Next, a sparse basis is used to improve DOA estimation. We implement the BSBL technique to determine the angle of arrival of multiple incident signals with only a single snapshot of data from an arbitrary arrangement of non-isotropic antennas. This is an improvement over the current state-of-the-art, where restrictions on the antenna type, configuration, and a priori knowledge of the number of signals are often assumed. Lastly, we investigate the feasibility of a basis set to reconstruct the scattering patterns of electrically smalltargets. The basis is derived from the TCM and can capture non-localized scattering behavior. Preliminary results indicate that this basis may be used in an interpolation and extrapolation scheme to generate scattering patterns over multiple angles.
Radar signalprocessing for maritime surveillance and reconnaissance is a difficult problem due to the occurrence of sea clutter. The sea surface is highly reflective and agile, therefore the received radar return has...
详细信息
ISBN:
(纸本)9781728116792
Radar signalprocessing for maritime surveillance and reconnaissance is a difficult problem due to the occurrence of sea clutter. The sea surface is highly reflective and agile, therefore the received radar return has a high intensity and it is often correlated in time and space due to undulation. Classical approaches try to suppress the clutter signal on the sensor or measurement level, however such methods usually do not take temporal information over multiple dwells into account. This article provides an analysis of an approach that simultaneously tracks the observed objects and the correlated clutter signals and classifies them based on the correspondence of their state history with the target- and clutter-specific dynamic models. Furthermore, it is demonstrated that this tracking based classifier maintains good accuracy even on downsampled data, which is representative of a scanning radar. The described method is tested on datasets of the 2006 Fynmeet trial conducted by the Council for Scientific and Industrial Research (CSIR).
In areas of complicated near-surface structures, estimation of an accurate near-surface velocity is challenging and this usually results in a distorted image of deeper targets. Surface waves have strong energy and dec...
详细信息
ISBN:
(纸本)9781613996751
In areas of complicated near-surface structures, estimation of an accurate near-surface velocity is challenging and this usually results in a distorted image of deeper targets. Surface waves have strong energy and decay exponentially with the depth, but they usually have high signal-to-noise ratios. Surface waves are strongly depending on S-wave velocities and also include information of P-wave velocity, density and attenuation of both P- and S-waves. Conventional inversion based on dispersion curves can only produce 1D S-wave velocity model. Full waveform inversion of surface waves breaks this limitation and can provide 2D or 3D high resolution near surface S-wave velocity, and reasonable P-wave velocity and density as well. Envelope of seismogram contains effective lower frequencies information which usually is not available in the seismogram. Combined with a multi-frequency strategy, cascadeded inversion of envelope-based and waveform-based misfit functions of both Rayleigh and body waves can reduce cycle skipping for FWI and estimate models with high resolution. This is due to the broader frequency bandwidth and composite contributions from Rayleigh waves and body waves. We test this method on the Arid 2D model, a typical near surface structure designed to include challenges in land seismic processing. Due to the high sensitivity to S-wave velocity and strong energy of Rayleigh waves, concurrent FWI of Rayleigh and body waves using cascaded envelope and waveform inversion produces S-wave velocity model with highest resolution. The resolution of P-wave velocity and density is lower than that of S-wave velocity. The small low velocity anomalies and faults at very shallow depths in the P- and S-wave velocity models are clearly defined. The low velocity structures at greater depths are better inverted in the S-wave velocity than in the P-wave velocity. Copyright 2020, International Petroleum Technology Conference.
暂无评论