Food security is going to be a problem and several efforts are being taken to help with it. Adoption of vertical farming, sensor based agriculture and GMCs. GMCs have been a part of us since TODO, the roma tomato which we consume everyday obtained a longer shelf life in 1984 by changing a gene. Great! genetically modified crops are helpful then what’s the problem? Finding the gene with the desired effect and no side effect is challenging job. Previously this was done by randomly mutating plants and observing the traits it exhibits. Currently, better methods are being applied to use the gift called DNA to its best.

The big picture

Imagine a future where droughts aren’t a problem, where plants can be grown very quickly with less resources, yields are maximum, farmer can be alerted with a problem before it can occur, etc.'Image based plant phenotyping' is one such emerging technique that helps us address this issue. Phenotypes are observable characteristics or traits of a plant. These traits generally reflect a combination of influences from the plant's genes and the environment it has grown and developed. Next let’s look at the term ‘image based’, if we use images to observe the variation in traits of plants then it’s image based process (what can be the other way of observing? Manually observing by measuring or destroying the plant). Now we can put it all together, using computer vision techniques to measure and analyze traits of plants of different genes to classify the effect of gene on the plant.

My contribution

My contribution to this problem was by determining and predicting when the plant came out of the soil. We call this emergence. Why is this important? This is helpful in farming where plant condition is assessed by a machine using camera. It acts as the early indicator of specifying something is wrong with a bunch of plants and enables farmers to re-plant them or fix the problem at hand rather than losing yield. This is also applicable in small-scale or personal farming. Apart from that, this is a contribution to plant phenotyping by allowing researchers to accurately study it in different conditions. Currently, a researcher monitors if the plant emerged or not and this can be become inscalable with several plants. It can become hard to track which plants were being monitor and their parameters.

Great! Let’s leave this job to the computers. This isn’t as easy as it sounds. It can be hard to find the plant in an image full of soil when you don’t know where to look and when to look. Shadows and illumination changes throughout the day can pose challenges too. Hence we use spatio-temporal technique and handle segmentation of plant during day-time and night-time as separate challenges. Lastly here are some of the results from the algorithm.

You can read the whole thesis if you are interested using this link. If you read till here, I am glad that you took time to read about the current problems the food industry faces and some ways to address it.

Recycling industry faces many challenges. Creating awareness and solving it from the grass root level is an important part of the initiative. As a volunteer in my university's recycling team, we used to walk around to every tailgate party and educate people about what they should recycle during their party. We also provided them with a recycle bag which made it easier for us to correctly recycle their disposed waste. With the power and reach of alexa based devices, this process can be much more streamlined to a service. My alexa skill, captain planet, aims to provide this service.

The idea

Instead of just providing facts or statistics about recycling items, I want to make and measure the impact of the alexa skill on the users. This can be done by dividing into three parts - engage, educate and impact.

Engage

By providing recycling facts to the users, we can engage them to find out more about how to recycle an item and the kind of activities they can perform to contribute to the recycling community.

Sample Utterance: Tell me some facts about recycling a soda can?
Sample Response: The energy you save by recycling a single aluminum can will run a TV for 3 hours. Would you like me to suggest tips about recycling soda can?

Educate

After engaging the user by providing the user some facts about recycling the desired item, we would like to educate them about some ways to correctly recycle that item.

We also ask the user about their interest in knowing about some recycling tips after the engagement step. The user can respond by saying responses like yes/no. This helps direct the user towards recycling methods instead of letting them figure out the questions to be asked.

Sample Utterance: Get me some tips on recycling soda cans.
Sample Response: It’s better to not crush soda cans as crushing it can get it mixed with paper by the processing equipment … it’s not a good thing, let’s not do it

Impact

Finally, we try to make an impact by suggesting user some ‘actions’. These actions can be finding nearby recycling services, buying eco-friendly items, joining nearby recycling events and organizations, etc. We can integrate APIs to provide this service.

Sample Utterance: Tell me about recycling bulbs
Sample Response: For several years, CFLS were considered as the best choice for eco-friendly consumers because they are much more energy efficient than typical light bulbs. But there is a big problem with CFL bulbs: they contain mercury, which is not good for the environment or human health. Consider using LED bulbs instead. I can add some LEDs bulbs to your card, would you like me to?

Implementation

Apart from the overall vision above, the implementation of the idea is important too and after a few iterations I have added some value to that too. The features are - internationalization, extendibility, measured impact, directed flow, room to encourage industry to be involved.

I made multiple attempts to obtain the API key for a service called Earth911 as they have a good collection of recycling info about several products. They also have information about nearby recycling services and the items they accept along with tips to recycle them. At the end, I couldn’t get the API key and decided to make a sample database of recycling info. This was divided into ‘facts’, ‘tips’ and ‘actions’. The source of these attributes also is stored as it is important to factor in the source of the information. We also store the item’s synonyms as comma separated keys to avoid redundant information and make it easier to update the recycling info.

The architecture of the application can be extended for several languages, as recycling is similar everywhere and language shouldn’t be a barrier. We also can different APIs to get the info about different aspects about recycling.

The application uses analytics platform called ‘voicelabs’, which helps us observe the impact of the skill by measure the users that invoke the different actions. This can also be used to determine the blockers that causes difficultly to the user for navigating through the skill.

Future Work

The actions like buying products and finding nearby organizations is not implemented yet as I was facing trouble with obtain device system information like address.

Creating bigger database about recycling by collaborating with organizations like Earth911.

I would also like to remember the facts, skills and actions that were suggested to the user and not repeat it in each session. We can also have some special dialogue regarding a product when matched with a synonym.

A tic-tac-toe bot for a little distraction during those mind-blockers. Play with computer or with a friend.

Demo