kAHOY Mate!: Mobile Application for Wood Identification using Deep Learning to Identify Illegally Logged Wood

Illegal logging continues to be a major threat to sustainable forestry in the Philippines. One effective way to combat this is by accurately and quickly identifying the different wood species of the confiscated wood. There are various ways to identify the wood species, such as traditional manual inspection, molecular-level research, and machine vision; however, not all of these methods are easily accessible to individuals in remote settings. With the lack of accessible wood identification techniques, kAHOY Mate! was created. This offline-functioning mobile application allows users to upload wood images, with or without an external macro lens, and classify them into four specific wood species and an ‘Others’ category. Deep learning architectures, specifically ShuffleNet V2 and MobileNet V2, were used to create a model for wood identification, achieving average validation accuracies of 95.57% and 76.03%, respectively. The MobileNet V2 model achieved a test accuracy of 66% and its TensorFlow Lite converted equivalent achieved an accuracy of 71%. The mobile application was developed using TensorFlow Lite and Flutter. This final mobile application achieved an accuracy of 50.50%, with an in-app camera accuracy of 29% and an image upload accuracy of 72%.