Image classifier with variable receptive fields in convolutional layers

摘要

Image classifier (1), comprising at least a first neural network (2), which is designed to an input image (10), and / or at least a portion (10a) of this input image (10), one or more classes ( 3a-3c) of a predetermined classification (3), the first neural network (2) having at least one convolution layer (21-23) which is designed to successively load at least one filter core (21a) at different open positions (11) applied to the input (10) of this convolutional layer (21-23) by summing values (12) of the input (10) in a weighted environment (11a') of the up position (11), the environment (11a') by supplying an environment (11a) proposed by the filter core (21a) to at least one image (4) whose behavior is characterized by image parameters (4a), and the number of image parameters (4a) being independent of is the size of the receptive field of the filter core (21a).Ver driving (100) for training and driving (200) with a complete chain of effects up to the control of a vehicle (50).