Morphological Shared-weight Neural Networks Research Articles

Machine Learning and Network Methods for Biology and Medicine

Machine Learning and Network Methods for Biology and Medicine

Vehicle detection from high-resolution satellite imagery using morphological shared-weight neural networks

High-resolution satellite imagery has recently become a new data source for extraction of small-scale objects such as vehicles. Very little vehicle detection research has been done using high-resolution satellite imagery where panchromatic band resolutions are presently in the range of 0.6–1.0 m. Given the limited spatial resolution, reliable vehicle detection can only be achieved by incorporating contextual information. Here, a GIS road vector map is used to constrain a vehicle detection system to road networks. We used a morphological shared-weight neural network (MSNN) to learn an implicit vehicle model and classify pixels into vehicles and non-vehicles. A vehicle image base library was built by collecting more than 300 cars manually from test images. Strategies to reduce the false alarms and select target centroids were designed. Experimental results indicate that the MSNN performed very well. The detection rate on both training and validation sites exceeded 85% with very few false alarms. By learning the implicit vehicle model through a MSNN, our method outperforms a baseline blob detection method.

Domain learning using Choquet integral-based morphological shared weight neural networks

Morphological shared weight networks (MSNNs) have been applied to many different problem domains, such as discriminating textures, target and hand written character recognition problems. MSNNs use morphological hit-and-miss transforms (HMT) as the activation functions in the feature extraction layer. However, conventional morphological filters have certain drawbacks. They are not robust and therefore are sensitive to noise. They may not tolerate small changes in gray values and shape. On the other hand, Choquet integral-based morphological operations (CMOs), a natural extension of classical morphological operations, are less sensitive to degradations in an image. The MSNN architecture is extended using CMOs in place of HMTs. These networks are referred to as Choquet morphological shared-weight neural networks (CMSNN). In a related paper, we compared the MSNN and CMSNN by extensive experimentation on the real world problem of land mine detection. In this paper, we discuss the problem of domain learning, which is related to feature importance. We provide examples of successes and failures of generalized mathematical morphology and the Choquet integral. These results are interesting, not only for purposes of robustness, but when viewed in the context of determining more valuable information sources.

Morphological regularization neural networks

In this paper we establish a relationship between regularization theory and morphological shared-weight neural networks (MSNN). We show that a certain class of morphological shared-weight neural networks with no hidden units can be viewed as regularization neural networks. This relationship is established by showing that this class of MSNNs are solutions of regularization problems. This requires deriving the Fourier transforms of the min and max operators. The Fourier transforms of min and max operators are derived using generalized functions because they are only defined in that sense.

LADAR target detection using morphological shared-weight neural networks

Morphological shared-weight neural networks (MSNN) combine the feature extraction capability of mathematical morphology with the function-mapping capability of neural networks in a single trainable architecture. The MSNN method has been previously demonstrated using a variety of imaging sensors, including TV, forward-looking infrared (FLIR) and synthetic aperture radar (SAR). In this paper, we provide experimental results with laser radar (LADAR). We present three sets of experiments. In the first set of experiments, we use the MSNN to detect different types of targets simultaneously. In the second set, we use the MSNN to detect only a particular type of target. In the third set, we test a novel scenario, referred to as the Sims scenario: we train the MSNN to recognize a particular type of target using very few examples. A detection rate of 86% with a reasonable number of false alarms was achieved in the first set of experiments and a detection rate of close to 100% with very few false alarms was achieved in the second and third sets of experiments. In all the experiments, a novel pre-processing method is used to create a pseudo-intensity images from the original LADAR range images.

Automatic target detection using entropy optimized shared-weight neural networks

Standard shared-weight neural networks previously demonstrated inferior performance to that of morphological shared-weight neural networks for automatic target detection. Empirical analysis showed that entropy measures of the features generated by the standard shared-weight neural networks were consistently lower than those generated by the morphological shared-weight neural networks. Based on this observation, an entropy maximization term was added to the standard shared-weight network objective function. In this paper, we present automatic target detection results for standard shared-weight neural networks trained with and without the added entropy term.

Entropy optimized morphological shared-weight neural networks

Morphological shared-weight neural networks previously demonstrated performance superior to that of MACE filters and standard shared-weight neural networks for target detection. Empirical analysis showed that entropy measures of the morphological shared-weight networks were consistently higher than those of the standard shared-weight neural networks. Based on this observation, an entropy maximization term was added to the morphological shared-weight network objective function. In this paper, target detection results are presented for morphological shared-weight networks trained with and without entropy terms.