Overview / Usage

Collect images and categorize them based on the deforestation in the area and plant growth/changes. Collected overtime, this data can show how urbanization and human behavior effects the plant life around us, showing changes in the types of plants that are growing, deforestation effects, and how plant life adapts overtime.

Image Source: https://software.intel.com/en-us/blogs/2018/09/07/rosemarie-day-wins-the-intel-ai-interplanetary-challenge

Methodology / Approach

Looking back at this issue, AI can be used to classify plants and analyze deforestation and growth to give a detailed view of how the world is changing over time. Using the Intel optimized TensorFlow framework, images of the earths surface can be taken by satellite and classified by location (country), crop/plant grown, and deforestation areas. This would need three classifiers to apply to images: (1) Location Classifier, (2) Plant Classifier, and (3) Deforestation Classifier.
 
Once a model for image classification has been created using TensorFlow, OpenVino can be run for model optimization on the Intel Nuc running Ubuntu. This creates the .xml and .bin files needed to run the inference engine on the user application. Once the model has been optimized, it can be run on the Intel Movidius Neural Compute Stick (NCS) hardware, also connected to the Nuc. Note: Another computer can be used in place of the Intel Nuc but needs to run a supported OS for software. The suggested on for this problem solution is a x86_64 computer running Ubuntu* 16.04.
 
The NCS allows for low-power applications that require real-time inferencing. By using the NCS, this allows for real-time monitoring of the earth's surface and visual awareness to be present in situations where power-constraints are more sever.
 
After detection reaches a certain threshold, images can be stored via the cloud for further analysis later. Such providers, such as AWS, have storage solutions for images. Images can then be broken down by region for storage. This allows scientists to go into a specific region to find images for further analysis into the plant life and deforestation effects.     

Combined with other largely available datasets for population growth and urbanization, scientists can use visual awareness and classification of the earths surface to better understand the effects and changes of human migration from rural to urban areas. Other data that can be analyzed along side images classified in this project would be weather data to look further at the effects of urbanization and deforestation on earths weather. Analyzed overtime, patterns could emerge based on the changes of weather compared to human migration.     
 
Steps for Proposed Solution
Train the model for image classification for plant type and country using TensorFlow.

Configure the OpenVino model optimizer for the TensorFlow model.

Freeze the TensorFlow model and convert this model in order to produce an optimized intermediate representation (IR) of the model. This model is based on the trained network topology of the TensorFlow model, the weights, and the bias values.

Test the results obtained from the IR format using the inference engine in OpenVino on the target environment for the application. The target environment is the Intel Movidius NCS.

Integrate the final model with the inference engine for the application into the targeted environment after testing.

Run model with threshold for detection. If results exceed a set threshold, push image to a cloud storage repository based on region image was taken in. For example, images of the US would be placed into a folder for the US. 

Technologies Used

Intel Optimized TensorFlow
OpenVino
Intel Movidius NCS
Intel Nuc running Ubuntu

Other links

• October 2018 | Intel® Developer Zone Update | Intel Software
• Rosemarie Day Wins the Intel® AI Interplanetary Challenge
• INTEL® AI Interplanetary Challenge Super Explorer Mission
• Intel announces winner of its AI Interplanetary Challenge
• Intel AI Interplanetary Challenge Winner Proposes Using AI for Earth-wide Plant, Deforestation Database