Mobile Based Mosquito Larvae Recognition from Photo Image Using Convolutional Neural Network

Abstract

The high number of dengue fever in Indonesia is a severe problem that affects the health of the Indonesian people. Coupled with the pandemic conditions that limit all movements of various groups due to social distancing, starting from health volunteers who must serve the community related to Covid-19. Routine inspection of mosquito larvae in every household must still be conducted to avoid the breeding of these mosquito larvae. Thus, an automatic and independent mosquito larva recognition system is needed from images taken via mobile devices that make it easier for each family head to identify areas of the home environment whether there are mosquito larvae or not. This paper proposes a new approach to larva recognition using convolutional neural network based on the TensorFlow library. The TensorFlow Serving and TensorFlow Lite were evaluated to get the best model with limited memory so that it can be used on mobile devices. The system classifies photo images into 2 labels, namely larvae or not. The result of this system is the presence or absence of mosquito larvae from the image. The TensorFlow Serving model produces an average testing accuracy of 96.1%, which is better than TensorFlow Lite of 90.1% in testing various conditions. The developed convolution neural network model produces a better model than the previous method.