Awards & Nominations
Spacia has received the following awards and nominations. Way to go!
Automated Detection of Hazards
Fire Force
Summary
Using Machine Intelligence and high-resolution data provided by NASA for quick detection of Wild Fires
How We Addressed This Challenge
We developed an ML/DL tool that uses Keras+TensorFlow backend which when given data according to the input parameters, is capable of predicting the presence of fires in the given area. Annually, delayed detection of fires causes the loss of several million acres of land, which may contain wildlife or even human civilization. Quick and effective detection of these fires plays a crucial role in mitigating their overall impact and aids the concerned authorities to address this disaster faster. Our model is a sequential one with 3 fully connected dense layers, having 32, 64 and 128 perceptrons respectively with 10,000 trainable parameters and dropout layers which 8 handpicked input parameters and then tries to predict whether there are fire pixels or not. We first train it on handpicked data acquired from the provided resources, and after the weights for the model have been established, it is capable of making accurate predictions on its own. With this project, we aim to enable firefighters and national authorities to have an even quicker response to natural disasters by harnessing the power of data science and machine learning. Our model will analyze the expansive database provided by NASA and aid the authorities in determining whether the given data corresponds to a fire or a false alarm. Aside from detection, our tool will be capable of alerting the authorities of civilizations that are in danger due to said fire.
How We Developed This Project
Our team chose this challenge because even though each of us belongs to different disciplines, data science is something we knew we could apply across our fields. We started with the project by browsing through the provided databases. After having a look at our data and deciding what we want to accomplish with it, we decided the structure of our model, and the input parameters we wished to train it on. We decided on Python as the code base with a Keras+TensorFlow DL model as the backend, and Streamlit+PyDeck frontend to make our tool interactive and intuitively accessible. Most of the project was developed as a team effort on Google Colab. The main problem our team had was with data collection, as none of us was familiar with the vast amount of platforms provided to gather data. We tried many of the listed databases but weren’t initially able to pinpoint which parameters in the data would be relevant to what we are trying to predict. Finding data from relevant satellites and figuring out how to parse, classify and ultimately use it effectively was nothing short of an achievement for all of us, and a great learning experience.
How We Used Space Agency Data in This Project
We examined the available features from various different datasets provided by NASA, and ultimately decided on the FIRMS dataset. We then performed manual feature engineering by studying the subject and decided on the following parameters: Land surface temp, aerosol_cldmax, aerosol_type land, aerosol_cloud_fraction_land, Deep_blue_aerosol_depth_550, combined, qaflag and algoflag. Finally, we trained a 3 layer deep affine model to make accurate predictions on whether coordinate is on fire or not.
Project Demo
Project Code
Data & Resources
We made use of the FIRMS Dataset to detect fires, and did manual feature engineering for the model to get accurate predictions.