Malaria is a life-threatening mosquito-borne blood disease, hence earlydetection is very crucial for health. The conventional method for the detectionis a microscopic examination of Giemsa-stained blood smears, which needs ahighly trained skilled technician. Automated classifications of differentstages of malaria still a challenging task, especially having poor sensitivityin detecting the early trophozoite and late trophozoite or schizont stage withlimited labelled datasize. The study aims to develop a fast, robust and fullyautomated system for the classification of different stages of malaria withlimited data size by using the pre-trained convolutional neural networks (CNNs)as a classifier and multi-wavelength to increase the sample size. We alsocompare our customized CNN with other well-known CNNs and shows that ournetwork have a comparable performance with less computational time. We believethat our proposed method can be applied to other limited labelled biologicaldatasets.