Our Prize Winning AWS Hackathon Entry > 자유게시판

• 목록
• 답변
• 글쓰기
• 게시판 리스트 옵션
  • 수정
  • 삭제

This article is written to elucidate my venture that I submitted to the AWS Hackathon on Devpost. Me and my buddy had initial started the undertaking as a learning step to build and deploy a machine learning challenge. We finally ended up successful the runner up for Computer Vision for our efforts. With rapid urbanization and industrialization Air pollution has turn out to be an alarming concern globally. Among different air pollutants particles , matter in (PM2.5) measurement vary are ready traverse deeply into our respiratory tract , transmitting hazardous chemicals into the human lungs and blood causing respiratory and cardiovascular well being points . We aim to develop image-primarily based air high quality evaluation utility , particularly a one that estimates the concentration of particulate matter with diameters in range of 2.5 micrometers. It will probably inform you what your breathing from photos generated using client grade hardware. Convolutional Neural Network (CNN) with the excessive computational power of Amazon EC2 DL1 instance to solve a computer imaginative and prescient problem of classifying natural pictures into 3 different classes ;wholesome , average and hazard based on their PM2.5 concentrations.

We created our own dataset by combining information from 2 sources . We discovered that, some causes for overfitting may very well be excessive imbalance in information , we had 2232 healthy images where because the average and BloodVitals review danger classes had only 1480 and 591 photographs respectively . 2232. Then we skilled our model following the identical methodology as before but this time we used ImageNet weights un-freezed the final one hundred layers for training , then wonderful-tuned the mannequin with a decrease learning charge ,we also used early stopping callback with a persistence of three . Now, there are some extra issues we've got completed to improve our mannequin. We migrated our mannequin onto the dl1 occasion and initially educated using 1 gaudi processor. This considerably improved our efficiency and efficiency .We had been capable of shortly take a look at , experiment and tweak adjustments . 76 %. We then transformed our model into a tensorflow lite model and used a huggingface demo software called gradio to quickly show our application.

Initially as we tried coaching our model we observed some of the image recordsdata in the dataset had been corrupted .We automated the technique of scanning by means of and deleting these information and then proceeded to training model . We used EfficientNet architecture since its renown for its performance in low spec machines, this known as switch studying . 30 layers and set the weights parameter to false. 60 % however our coaching accuracy was high, the model may very well be overfitting. As now we have seen we have been able an accuracy of about 76 % which is fairly good considering the data we had. Now , why do we want this application and how is it different from the prevailing air monitoring system systems, we already know Exposure to wonderful particles could cause long run well being results resembling lung and coronary heart drawback but they also cause short time period well being results akin to eye, nose, throat irritation.

The existing air high quality monitoring methods are extremely dependent on monitoring stations, that are once more situated far away because of the high setup price and costly equipment. For a selected location, the air high quality depends on how air strikes via the world and how people are influencing the air , so we can't rely on something that was predicted a while back or in some place that is not our site of curiosity .This application helps you analyse the air high quality around you with a single image and extra importantly is portable. We are able to carry it wherever we wish to and know what your being exposed to . Honestly I don’t know, I believe there may be potential for this app to integrate into our routine. People would possibly want options to assist plan their day out, integrate with their phone’s calender and BloodVitals review recommend the very best time to plan an out of doors activity. Right now the Artificial intelligence that runs the machine runs on the azure cloud. Sooner or later we wish to have the ability to carry it into the app and run it natively.

Certain constituents within the blood affect the absorption of light at various wavelengths by the blood. Oxyhemoglobin absorbs light extra strongly in the infrared area than in the purple area, whereas hemoglobin exhibits the reverse conduct. Therefore, highly oxygenated blood with a high focus of oxyhemoglobin and a low concentration of hemoglobin will tend to have a excessive ratio of optical transmissivity in the crimson region to optical transmissivity within the infrared area. These alternating portions are amplified after which segregated by sampling gadgets working in synchronism with the red/infrared switching, so as to supply separate indicators on separate channels representing the crimson and infrared light transmission of the body construction. After low-pass filtering to remove signal components at or above the switching frequency, each of the separate signals represents a plot of optical transmissivity of the physique construction at a particular wavelength versus time. AC element brought about only by optical absorption by the blood and various on the pulse frequency or coronary heart rate of the organism.