I have always been a fan of agriculture and of organic food. It is a field dedicated to the everyday survival of the human race.
Lack of indigenous technological solution to Nigeria's addressable agricultural problems.
Amidst the variety of problems plaguing the agricultural industry of Nigeria, we have chosen to focus on a few these are at the moment --Plant Disease Detection --Plant Pest Detection --Animal Disease Detection --Soil Fertility Level Detection and Crop Recommendations. The plant disease reduces crop yield and Nigeria is a country which is trying to bring the agricultural sector to the glory it once had, not to mention it's exploding population. Crop yield reduction can have massive impacts on feeding, because less people are feed, the available food become costly and unaffordable.
Do existing solutions for some components of this projects exist? The answer is yes, I personally believe that no idea is completely brand new, all you do is improve on its current state. The current solutions to problems like this similar to our idea is the use of a trained neural network model placed on a server which several devices on several platforms can connect to. The downsides to the current solutions are: --It requires internet connection between the client side and the server side. --Although APIs are designed to make data transfer from source to destination seamless there is still the issue of network latency and in the case where multiple people are accessing the same platform at the same time, which will definitely happen at scale, a queue will begin to occur which will increase processing time, this implies that a process that would take 5 seconds to complete would now be taking almost a minute if I had just 10 people on the queue before me. This kind of solutions are centralized in nature.
Our aim is to bring the ability to detect all the above listed things on your mobile phone by placing the neural networks needed to carry out these operations as a component of the offline applications developed for mobile, i.e iOS and Android.
How it was built
One of the very first things we did was design the template for the software. Then we worked towards this template. We used technologies such as TensorFlow lite, which is a scaled down version big the TensorFlow library used in neural network training but in this case it allows us to use a scaled down version big our trained model on the mobile devices. We also used technologies such as java, Android studio to develop the Android Application as well as software like Xcode in developing the iOS application.
Challenges We Ran Into
Giving the fact that neither of team members are developers for mobile, one of the major problem we had was learning how it should be done and developing an application at the production level, within the time frame. We also needed lots of internet subscription, as a single file such as Xcode had a file size of 7.3GB, which is very large considering the available network subscriptions. Another problem we faced is a source to get good enough data to train a neural network.
--Our developed software will be easier to access. --Our developed software will be faster. --The neural network will be decentralized. --We can deploy a single neural network to carry out multiple tasks. --Our developed software will run completely offline! Hence, they'd be no issue of lack of internet connectivity.
Why trust our project?
Our team comprises of two people, I Akinsola Adefolahan and Steven Kolawole. I am a Machine learning developer, and I've had experience developing Deep Learning solutions to problems. My field of study is Mechatronics Engineering, my university of study is the Federal University of Agriculture, Abeokuta and as such I have a form of exposure to agriculture. I also have a personal experience with agriculture in terms of a family farm.
One of the major accomplishments for me in this project is the fact that we were able to complete the task, although unforeseen circumstances were encountered. Another major accomplishment is the fact that every individual member of the team was able to work together and augment each others effort.
What I learned
I have never worked in the field of Android or iOS apps development, but due to sudden situations which is actually the fact that a project similar to this was done in NaijaHacks2018 we had to come up with immediate ways of improvising. So we decided to go a step further by making the entire application offline and removing the the need for internet access, the server-client system before accessing the neural network for use. This taught us that we should always be ready to make changes to our original plans, as plans don't always work as planned, adaptability is important. We were able to gain even though little, but knowledge about android and ios development through the use of JAVA, Android Studio, Swift, Xcode.
Our future plans
One of the most important feature we want to implement is to make sure that this software is deploy-able across multiple platforms.
There is the possibility of incorporation of IOT projects in order to facilitate the swift response of the farmer to detected problems such as an automated dispensing of pesticides if a pest is detected, with the approval of the farmer, but requiring minimal effort.
Our apps are not currently available on the store bu tcan be downloaded here