Leukemia is a critical condition affecting the blood and lymphatic systems, requiring timely and accurate detection for effective treatment. Traditional microscopic analysis, though effective, relies heavily on the skill of the pathologist, leading to potential delays in diagnosis. In this research, a novel introduced LeukNet proposed for identifying the different types of leukaemia in its early stages. Initially, the images of the blood smear are preprocessed using Adaptive Gaussian filters to removing noise artifacts. Sobel detector is used to detecting the edges in horizontal and Vertical from the images. Based on the edges, the deep learning-based GoogleNet is used for extracting the blood cell features. Further, the classification is performed by Multi-Layer Perceptron. It classifies the smear of blood images into five different classes: Normal, Acute leukemia (ALL), Acute lymphoblastic leukemia (ALL), Chronic lymphoblastic leukemia (CLL), and Chronic myeloid leukemia (CML). The effectiveness of the proposed LeukNet method using metrics like F1 score, sensitivity, accuracy, and specificity. The proposed LeukNet model achieved a classification accuracy 98.76%. The proposed LeukNet model enhanced the total accuracy by 2.90%, 6.57%, and 8.48% better than LD-C NMC, ALL-delt and DeepLeukNet respectively.