Malaria is a potentially fatal, yet preventable and curable disease transmitted to humans by certain mosquito species, particularly in tropical regions. Five Plasmodium species can cause malaria in humans, with P. falciparum and P. vivax posing the greatest threat. The gold standard for diagnosis remains the microscopic examination of stained thick blood smears; however, this method depends heavily on analyst expertise and local resources. In this study, we developed a convolutional neural network (CNN) model based on MobileNetV2, a 53-layer architecture pretrained on over one million ImageNet images. A quadrant-based image preprocessing strategy was used to enhance parasite localization and reduce overfitting. The model was trained on a balanced dataset comprising healthy, P. falciparum, and P. vivax samples, transforming it into a multiclass classifier. The proposed model achieved a weighted F1-score of 97.7%, outperforming several recent state-of-the-art approaches. Unlike binary detection systems that only distinguish between healthy and infected samples, our multiclass model accurately identifies specific Plasmodium species, which are clinically relevant for detecting mixed infections and supporting appropriate treatment decisions. Thanks to its low computational requirements and high accuracy, this approach shows promise for integration into mobile diagnostic tools in resource-limited healthcare environments.