Neural network learning process and main tasks

The learning process by the software uses the training image dataset to define the values of the coefficients and the relative weights of the combination of vectors that form the neural network.

This learning process allows the software to create the network model, and it is usually an iterative process of updating neural network weights based on the training dataset. Using machine learning software is typically divided into two main stages:

First is the training stage, wherein the software generates a model based on features learned from training samples.

Then comes the inference stage, where the model is applied on new images to perform the actual machine vision task.

Since the number of coefficients and weights to be assigned depends on the Network Depth, the training complexity and the required computational performance of the hardware increase as the Network Depth increases.

Common tasks that can be solved by neural networks are:

• feature detection
• object classification
• instance segmentation
• object location
• reading characters