Lamokollect: A convolutional neural network (CNN) classification model for mosquitoes of medical importance

Malaria is one of the most life-threatening diseases transmitted to people through the bites of female mosquitoes of the genus Anopheles that were infected with protozoan parasites belonging to the genus Plasmodium. Anopheles mosquitoes are considered intermediate hosts or vectors of malaria in humans, which to this day still lacks preventive measures in controlling the growth of vector mosquitoes and identification of their species. This problem is being addressed in this study considering that mosquitoes from the genus Anopheles are of medical importance and could be used as a model in classifying certain species of Anopheles mosquitoes that are vectors of the malaria disease in humans through a novel species identification model. Convolutional Neural Networks (CNNs) are designed for image and pattern recognition, using hierarchical layers of convolutions to automatically learn and extract relevant features from input data. The input data is an image database composed of 957 images from six Anopheles species, namely Anopheles coustani, Anopheles crucians, Anopheles funestus, Anopheles gambiae, Anopheles punctipennis, and Anopheles quadrimaculatus, which have significant physical damage in many specimens, reflecting real-world conditions. To address the fine-grained morphological diversity and high intra- and inter-species variation, seven CNN models were employed and evaluated using transfer learning. Performance was assessed based on accuracy, loading time, and model weight size. GoogleNet achieved the highest accuracy of 95%, while YOLOv8 was the fastest and most lightweight model in terms of loading time and weight size, respectively. These findings demonstrate the potential of deep learning for mosquito species classification even with challenging image data, paving the way for automated vector identification in public health applications.