Aselsan Elektronik Sanayi Ve Ticaret Anonim Sirketi

Erdem Akagunduz

Tolga Aksoy

Huseyin Seckin Demir

Engin Uzun

Disclosed is a method for training shallow convolutional neural networks for infrared target detection using a two-phase learning strategy that can converge to satisfactory detection performance, even with scale-invariance capability. In the first step, the aim is to ensure that only filters in the convolutional layer produce semantic features that serve the problem of target detection. L2-norm (Euclidian norm) is used as loss function for the stable training of semantic filters obtained from the convolutional layers. In the next step, only the decision layers are trained by transferring the weight values in the convolutional layers completely and freezing the learning rate. In this step, unlike the first, the L1-norm (mean-absolute-deviation) loss function is used.

Method for training shallow convolutional neural networks for infrared target detection using a two-phase learning strategy

A filter design method for a small target detection on infrared imagery using a normalized-cross-correlation layer in neural networks, including the steps of: Normalizing inputs and filters of a convolutional neural network, wherein normalizing inputs and filters of the convolutional neural network provides faster convergence in a limited database. Defining a forward function of a normalization layer in the convolutional neural network, wherein the forward function of the normalization layer in the convolutional neural network is used for training a neural network. Defining a derivative function of the normalization layer for a back propagation in a neural network training phase. Training created neural networks with datasets, wherein the datasets consist of target and background views and using trained neural networks in the small target detection.

Filter design for small target detection on infrared imagery using normalized-cross-correlation layer in neural networks

The present invention relates to the field of image processing and methodologies to construct a descriptor from binary silhouette images. The method comprises the steps of; receiving a parametric equation of a closed planar curve, choosing nodes on the closed planar curve with equal intervals, generating a continuous scale space of the nodes on the curve by successively convolving the planar curve with a Gaussian filter and down-sampling it after each octave, calculating orientation vectors and orientation angle values for each sampled point at each interval of each octave, creating a orientation scale space (OSS) which is a matrix obtained from stacking each orientation angle value on top of each other, representing the outline with a descriptor including all the orientation vectors and their respective parameters position and scale.

System and method for describing image outlines

The present invention relates to a system and method for identifying scale invariant features of image outlines. The method comprises the steps of; receiving a parametric equation of a closed planar curve; choosing nodes on the closed planar curve with equal intervals; generating a continuous scale space of the nodes on the curve; calculating circle of curvature for every node on the closed curve for every scale in every octave; finding circle of curvature differences between plurality of adjacent scales; comparing each curvature difference value and choosing the nodes with a minimum or maximum curvature difference as feature points; representing the outline with a descriptor including all the feature points. The method further comprises the steps; eliminating the feature points which are closer to each other than a predetermined threshold; and comparing a descriptor with each previously recorded descriptor belonging to various outlines, finding at least one descriptor with a good match.

Growing community of inventors

Ankara, Turkey

Erdem Akagunduz