Konvolutionskärna

Konvolutionskärna, also known as a filter or kernel, is a fundamental concept in convolutional neural networks (CNNs). It is a small matrix of weights that is applied to an input image or feature map. The process of applying the kernel to the input is called convolution.

During convolution, the kernel slides over the input data, performing element-wise multiplication between the kernel's weights and the corresponding input values. The sum of these products forms a single value in the output feature map. This operation is repeated for every position of the kernel across the entire input.

The primary purpose of a konvolutionskärna is to detect specific features within the input data. For instance, a kernel might be designed to identify horizontal edges, vertical edges, curves, or more complex patterns. By using multiple kernels, a CNN can learn to recognize a hierarchy of features, starting from simple ones in the early layers and progressing to more complex representations in deeper layers.

The weights within a konvolutionskärna are not pre-defined; they are learned through the training process of the neural network. During backpropagation, the network adjusts these weights to minimize errors, enabling the kernels to become increasingly adept at detecting relevant features for the given task, such as image classification or object detection. The size and shape of the kernel are hyper-parameters that can be chosen by the designer of the CNN architecture.