Grain Analysis (Artificial Intelligence-Computer Vision)
Budget: $250 – $750 USD
We need to implement a Computer Vision (CV) system that it will receive images of grain trucks
unloading their cargo. With the cargo on the floor, it should identify the contents of the truck
(cereals).
To do this, a camera will be installed (camera will hold always the same position and angle),
which allows real-time images of the download to be taken and can classify the content into
one of 9 categories. (Categories are detailed below)
The solution must be implemented on Python. As computing requirements are not intensive,
we are considering using standard CPU and not GPU processing. We are not going to process
more than an image per minute. Image processing and output should not last more than 15
seconds.
We can consider as starting point a pre-trained neural network and use transfer learning
techniques to perform custom training of the model to achieve accuracy.
At this point, we do not have yet the training data set. Our hypothesis is that we will get no less
500 to 600 images in each of the classes we need to predict, maybe more.
Your deliverable should be a Python Script running on the server we have chosen (between you
and our team), with a technical document stating all the information on the model, training
methos, hyperparameters chosen, activation function, model accuracy, etc. etc.
The server will be mounted and configured by our team with your supervision and assistance. It
will most probably be a Linux type server on a i5 or i7 processor last generation, 16 Gb RAM,
1TB SSD HDD. (should we use an NVIDIA card?). Computer will be allocated only for this project.
Other Specs:
- Image Processing Threshold (on production): < 15secs
- Original Image Size: 1920x1080 RGB
- Training Images:
Please state on your proposal:
- All what you think you might need for completing the task.
- Risks and concerns
- Project Value
- Delivery Time
Find examples on attached document.
unloading their cargo. With the cargo on the floor, it should identify the contents of the truck
(cereals).
To do this, a camera will be installed (camera will hold always the same position and angle),
which allows real-time images of the download to be taken and can classify the content into
one of 9 categories. (Categories are detailed below)
The solution must be implemented on Python. As computing requirements are not intensive,
we are considering using standard CPU and not GPU processing. We are not going to process
more than an image per minute. Image processing and output should not last more than 15
seconds.
We can consider as starting point a pre-trained neural network and use transfer learning
techniques to perform custom training of the model to achieve accuracy.
At this point, we do not have yet the training data set. Our hypothesis is that we will get no less
500 to 600 images in each of the classes we need to predict, maybe more.
Your deliverable should be a Python Script running on the server we have chosen (between you
and our team), with a technical document stating all the information on the model, training
methos, hyperparameters chosen, activation function, model accuracy, etc. etc.
The server will be mounted and configured by our team with your supervision and assistance. It
will most probably be a Linux type server on a i5 or i7 processor last generation, 16 Gb RAM,
1TB SSD HDD. (should we use an NVIDIA card?). Computer will be allocated only for this project.
Other Specs:
- Image Processing Threshold (on production): < 15secs
- Original Image Size: 1920x1080 RGB
- Training Images:
Please state on your proposal:
- All what you think you might need for completing the task.
- Risks and concerns
- Project Value
- Delivery Time
Find examples on attached document.