The specific application tackled in this example, with Alexnet, is the winning solution to the Hackathon: https://blogs.itility.nl/en/image-recognition-model-that-identifies-plant-species. Use of a GPU is highly recommended.
Experts create and train Deep Neural Networks on millions of images to automate tasks like Object Recognition. See more: https://nl.mathworks.com/help/deeplearning/ug/pretrained-convolutional-neural-networks.html.
Such pre-trained networks are easily accessible and tunable with MATLAB to efficiently solve new problems.
Transferring the learning from those networks to a new problem is a simple operation on the network layers, that typically achieves good results with fewer images and less training time.
In this script, you can easily try out many of these pre-trained networks and explore: training options, compare accuracy results and find the settings that will give you the best 'Hackathon' results!
Once your network is ready you can save it as a *.mat file and test it live (on streaming video) using a webcam and the 'ClassifyImagesFromWebcam(net)' function.
You can also speed up the prediction of the model using MEX or deploy it directly on embedded devices with automatic code generation for GPU, CPU, etc.
Paola Jaramillo (2019). Deep Learning Hackathon with Transfer Learning (https://www.mathworks.com/matlabcentral/fileexchange/68328-deep-learning-hackathon-with-transfer-learning), MATLAB Central File Exchange. Retrieved .
Including more network architectures.