Charles Tian¹ and Yu Sun², ¹USA, ²California State Polytechnic University, USA

Birds are everywhere around us and are easy to spot. However, for many beginner birders, identifying the birds is a hard task [8]. There are many apps that help the birder to identify the birds, but they are often too complicated and require good internet to give a result. A better app is needed so that birders can identify birds while not depending on internet connection. My app, AI_Bider, is mainly built in android studio using flutter and firebase, and the AI engine is coded with TensorFlow and trained with images from the internet [9]. To test my AI engine, I made six different prototypes, each having a different number of times that the code will train from the dataset of pictures. I then selected 5 birds that are in my dataset and found 5 pictures on the internet for each of them, which I then uploaded to the app. My app will then give me 3 bird species that most closely resemble the image, as well as the app’s confidence in its choices, which are listed as percentages. I recorded down the percentages of accuracy for each picture. After taking the average percentage of all the models, I selected the most successful model, which had an average percent of accuracy of 79%.

Machine Learning, AI platform, Computer vision