Deep Convolutional Network Transfer learned from alexnet into LionFishSlayerNet
Training Loss of LionFishSlayerNet, the CNN becomes more accurate over minibatch training epochs. Training loss approaches 0 as t->∞
Hack river lagoon LionFishSlayerNet (We make high tech lionfish extermination tools)
Link to hackathon notes
How it works
Our project makes it easy to automate the identification and tracking of invasive species in the IRL.
In order to achieve this, we modified layers of a MATLAB CNN called alexnet and retrained with 1400 lionfish images.
Our deep learning software is called LionFishSlayerNet. Yeah, it's dope!
A Raspberry Pi is used as a target for the CNN.
A camera module connected to the Pi captures images that are fed to LionFishSlayerNet.
LionFishSlayerNet makes a prediction on the image. if it thinks it's a lionfish, it will stop and send data to an IoT cloud over wifi in real time.
The data is available here: https://thingspeak.com/channels/618366.
The data tells you, when the image was taken, where it was taken, and how confident the neural net is that it's a LionFish.
Boom! A.I and IoT do all the work while we sit back and relax.
LionFishSlayerNet can be trained to identify any other invasive species if you have a big enough dataset. The possibilities are endless!
Deep learning matlab support package Alexnet matlab package Raspberry Pi matlab hardware support package
Loss in accuracy when dealing with non-lionfishes Requires proximity to subject