COVER STORY u ANALOG DEVICES
Figure 4: Feature maps for a CNN
end, it forms, together with the convolutions, the unique properties of the respective object class. However, the result of these convolution and pooling groups is a large number of two- dimensional matrices. To achieve our actual goal of classification, we convert the two- dimensional data to a long one-dimensional vector. The conversion is done in a so-called flattening layer, which is followed by one or two fully connected layers. The neurons in the last two layer types are
www.irish-manufacturing.com
similar to the structure shown in Figure 2. The last layer of our neural network has exactly as many outputs as there are classes to be distinguished. In addition, in the last layer, the data are also normalised to yield a probability distribution (97.5% cat, 2.1% leopard, 0.4% tiger, etc.).
This concludes the modeling of our neural
network. However, the weights and contents of the kernel and filter matrices are still unknown and must be determined through network training in order for the model to work. This
will be explained in a subsequent article as will the hardware implementation for the neural network we have discussed (for cat recognition, as an example). For this, we will use the MAX78000 artificial intelligence microcontroller with a hardware- based CNN accelerator developed by Analog Devices.
Analog Devices
www.analog.com
Irish Manufacturing October 2023 13
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38 |
Page 39 |
Page 40 |
Page 41 |
Page 42 |
Page 43 |
Page 44 |
Page 45 |
Page 46 |
Page 47 |
Page 48 |
Page 49 |
Page 50
