Benyarut Erawan, Kittipop Choknitipakin, Poom Wettayakorn, Siripong Traivijitkhun

Initially, we retrieved GSV images and date of each image in specific sub-district of Bangkok in 2 different ways, along roads and grid boxes, by using Google API. We also retrieve Terrapattern’s satellite images from online source to see the performance of detecting water which could be the breeding site. After gathering images, we use TensorFlow image net to run the images, detect water and objects in the GSV images. Moreover, we use Caffee-Segnet, a Deep Convolutional Encoder-Decoder Architecture for Semantic Pixel-Wise Labelling, to classify GSV image into parts for further data analysis process to find mosquito’s breeding site from GSV and decrease the wrong detecting problem. On Monday 3rd July to Monday 10th July 2017, we prepare ground truth data for train and create an appropriate model on the mosquito’s breeding site detection by these following steps:

• Retrieve objects images from online source, ex. tire, spirit house, etc.
• Retrieve random GSV images from different areas in Bangkok, 30,000 images
• Pass all 30,000 GSV images through Caffe-Segnet model
• Pass all 30,000 GSV images through Tensorflow model

For our next step, we will prepare ground truth data for train and create an appropriate model on the mosquito’s breeding site detection by manually label all classified images from Tensor model, and also train new model with the labeled images and the objects images.