Odei Garcia-Garin et al. from the University of Barcelona have developed a deep learning-based algorithm able to detect and quantify floating garbage from aerial images. They also made a web-oriented application allowing users to identify the garbage known as floating marine macro-litter, or FMML, within images of the sea surface. Watch the video: References Odei Garcia-Garin et al., Automatic detection and quantification of floating marine macro-litter in aerial images: Introducing a novel deep learning approach connected to a web application in R, Environmental Pollution, . https://doi.org/10.1016/j.envpol.2021... Code & web app: https://github.com/amonleong/MARLIT Follow me for more AI content: Instagram: https://www.instagram.com/whats_ai/ LinkedIn: https://www.linkedin.com/in/whats-ai/ Twitter: https://twitter.com/Whats_AI Facebook: https://www.facebook.com/whats.artifi... Medium: https://whats-ai.medium.com/ Join Our Discord channel, Learn AI Together: https://discord.gg/learnaitogether The best courses in AI & Guide+Repository on how to start: https://www.omologapps.com/whats-ai https://github.com/louisfb01/start-ma... Become a member of my YouTube community: https://www.youtube.com/channel/UCUzG... Chapters: 0:00 - Hey! Tap the Thumbs Up button and Subscribe. You'll learn a lot of cool stuff, I promise. 0:30 - Floating marine macro-litter 2:19 - The method 5:10 - Conclusion Video Transcript 00:00 an ai software able to detect and count 00:03 plastic waste in the ocean 00:04 using ariel images it's both clever and 00:07 simple 00:08 and you could use this same model for 00:10 many image classification applications 00:12 let's see how it works 00:16 [Music] 00:21 this is what's ai and i share artificial 00:23 intelligence news every week 00:25 if you are new to the channel and want 00:26 to stay up to date please consider 00:28 subscribing to not miss any further news 00:31 we live on a blue planet over 70 00:35 of earth is covered by sea from space 00:38 our ocean appears pristine clean 00:42 unfortunately it's not because of poorly 00:45 controlled waste sites 00:46 illegal dumping and mishandled waste 00:49 from population centres 00:51 tourism industrial and agricultural 00:53 activities 00:54 an estimated 8 million metric tons of 00:57 plastic 00:58 waste entered the oceans 01:02 aude garcia garion et al from the 01:04 university of barcelona 01:06 have developed a deep learning based 01:08 algorithm able to detect and quantify 01:10 floating garbage from aerial images 01:13 they also made a web-oriented 01:15 application allowing users to identify 01:17 these garbages 01:18 called floating marine microliter or fml 01:21 within 01:22 images of the sea surface floating 01:25 marine macro litter is any persistent 01:27 manufactured or processed solid material 01:30 lost or abandoned in a marine 01:32 compartment as you most certainly know 01:34 these plastic wastes are dangerous for 01:36 fish turtles and marine mammals as they 01:39 can either 01:40 ingest them or get entangled and hurt 01:42 traditional approaches to detecting 01:44 these 01:45 fmls are observer-based methods 01:48 meaning that they require someone on a 01:50 vessel or airplane to look for them 01:52 yielding to precise identification but 01:55 extremely expensive and time demanding 01:57 labor 01:58 fortunately this detection can be done 02:00 using cameras or sensors on aerial 02:02 vehicles 02:03 but it also requires trained scientists 02:05 to manually look at the collected data 02:08 being again extremely time consuming 02:10 automation is clearly needed here and 02:12 could help us 02:13 improve the quality of our marine 02:15 compartments worldwide 02:16 much more effectively this is where 02:19 machine learning 02:20 and deep learning commonly deep learning 02:23 proves over and over 02:24 that it's a very powerful automation 02:26 tool and especially in the computer 02:28 vision industry 02:29 where it's known to automatically 02:31 identify the important features of an 02:33 image 02:33 without any human supervision making 02:36 this approach 02:37 less time demanding than its 02:38 predecessors for many different 02:40 applications including 02:41 this very important one as you may 02:44 suspect 02:45 they use the convolutional neural 02:46 networks to attack this problem 02:48 this type of neural network is the most 02:50 commonly used deep learning architecture 02:52 in computer vision 02:54 the idea behind this deep neural network 02:56 architecture is to mimic the human's 02:58 visual system if you want to learn more 03:00 about the foundation of convolutional 03:02 neural networks 03:03 or cnns i will refer you to their video 03:05 on the top right corner on your screen 03:07 where i'm explaining them more in depth 03:11 they train their algorithm with real 03:13 images like this one 03:15 taken by drones and aircraft with 03:17 annotations made by the same 03:19 professionals that are usually 03:20 manually analyzing them this is a 03:23 challenging task even for deep learning 03:25 because of all the possible variations 03:27 in colors and sun reflections as you can 03:29 see here 03:31 in short their model is a regular binary 03:33 classifier 03:34 cnn architecture composed of 03:36 convolutions and poolings 03:38 terms that i explained in the video i 03:40 referenced earlier 03:41 that outputs a binary response telling 03:43 us if there are fmls or not in the 03:46 picture 03:46 the depth of the network is due to these 03:48 convolution layers 03:50 compressing the image and creating many 03:52 feature maps 03:53 which are the outputs of the filters 03:55 ending with a general representation 03:57 of the image allowing us to know in 03:59 general 04:00 what the image contains such as fml in 04:03 this case 04:04 note that this exact same architecture 04:06 could have been used on 04:08 any other computer vision application 04:10 with a test to classify whether or not 04:12 something is in the image such as 04:14 putting a defect on a manufacturer part 04:16 or telling if there is a dog or not what 04:18 they did differently making it powerful 04:20 to fml detection 04:22 is that they had the idea to split the 04:24 image into 25 smaller cells 04:26 that each outputs a classification 04:28 result fml 04:30 or not yielding much better overall 04:32 accuracy 04:33 then they used the shiny package of r 04:37 to develop their application their 04:39 algorithm allows the detection and 04:41 quantification 04:42 of fmls as well as providing support to 04:45 the monitoring and assessment of this 04:47 environmental threat 04:48 however it's still not completely 04:50 automated yet and requires a human in 04:52 the loop 04:53 as of now they are still looking for 04:55 more annotated data to allow their 04:57 algorithm to also 04:58 identify the size color and type of fml 05:01 which are very relevant information for 05:03 planning well-targeted policy and 05:05 mitigation measures 05:07 this is still an amazing application of 05:09 deep learning with a great use case that 05:11 will benefit everyone 05:13 of course this was just an introduction 05:15 to this new paper 05:16 and i linked both the paper their code 05:18 and their application 05:20 in the description below if you would 05:21 like to read more about it or even 05:23 try it out yourself please leave a like 05:26 if you went this far in the video 05:28 and since there are over 80 percent of 05:30 you guys that are not subscribed yet 05:32 please consider subscribing to the 05:33 channel to not miss any further news 05:36 thank you for watching 05:40 [Music]
