Published Paper Details:

To enhance the accuracy of Aedes larvae identification stands as a crucial initial measure in combating mosquito-borne diseases such as Dengue and the Zika virus. This project proposes an innovative approach by harnessing the power of deep learning, specifically employing convolutional neural networks (CNNs), to automate the segmentation of Aedes larvae from images captured in diverse watery settings. Uploading pictures of prospective breeding locations is requested of users. After analyzing these photos, the deep learning network generates conclusions on whether Aedes larvae are present. Our proposed system comprises two integral components: Classification and Image acquisition & Segmentation. Classification: In our approach to mosquito larvae classification, we use deep learning architectures like VGG16, VGG19, ResNet50, ResNet152, and InceptionV3. The utilization of VGG16 and VGG19 enables us to harness deep convolutional networks with multiple layers, while ResNet50 and ResNet152 leverage residual learning to enhance model training and facilitate more effective feature extraction. Additionally, InceptionV3's inception modules contribute to the overall efficiency by capturing intricate spatial hierarchies in the larval images. Image acquisition & Segmentation: In our segmentation framework for mosquito larvae analysis, we adopt the Unet++ module, a superior alternative to the Unet segmentation approach. The decision to employ Unet++ stems from its demonstrated efficiency and notable advantages over the standard Unet model. The capabilities of Unet++ not only elevate segmentation accuracy but also contribute to improved boundary delineation and finer-grained analysis. By embracing Unet++, our segmentation module stands poised to deliver superior results, underlining our commitment to leveraging cutting-edge methodologies for robust mosquito larvae analysis in vector control efforts.