Hi, this is a Dust Detector Neural Network. Solar panels are getting more and more popular; however, many people don't know that when solar panels aren't cleaned they can lose up to 35% of their effectiveness. To help people pay less for electricity bills and contribute to the elimination of climate change I created this project.

THe first step was downloading a database which has images of both dirty and clean solar panels. Next step was to split it to TRAIN/TEST/VAL directories in order to train a model based on them. After spliting I deleted some images that were advertisements of solar panels. After spliting and removing the ads I started coding the main part of this project. To make my project more usefull and better instead of creating my own neural network I used a pretrained RESNET-18. I trained it for 40 epochs and increased the batch size to 12, instead of default 8. The model has around 72% accuracy and is the best one I could train. In the future I will try to make it more accurate and code my own neural network from scratch.

1. Make sure you downloaded jetson-inference library and built it succesfully
2. Download the folder of model Solar2 or any model in models folder
3. Download modified version of imagenet.py file
4. Remove imagenet from jetson_inference/build/aarch64/bin/ and place the new one in it
5. Place your image jetson-inference/python/training/classification directory
6. Download runALL.py file
7. Run runALL.py --name --model command with the file name of the image and the model you downloaded
8. Your result will be a <name>__RESULT.jpg

[View a video explanation here](video link)