Due to the complex background information and too much interference information of the optical melanoma image collected by dermoscopy, the detection accuracy is low, and there are problems such as false and missed detection. Therefore, an optical melanoma image detection algorithm based on reparameterized large kernel convolution is proposed. Firstly, in the backbone part, a new module C3_RepLK that combines large kernel convolution and C3 was designed to increase the receptive field of the model and extract more effective information. Secondly, the receptive field module RFB was introduced to fuse the feature information of different scales to reduce the problem of false detection. The hybrid dense sparse convolution GSConv and lightweight upsampling operator CARAFE were used in the neck network, so that the network could capture rich context information and suppress the missed detection problem. Finally, the second-order channel attention module SOCA was integrated into the algorithm to strengthen the correlation between features and focus on more useful features. Experimental results show that compared with the original YOLOv5 algorithm, the proposed detection algorithm improves the average accuracy of all categories from 85.0% to 89.4%, which proves the effectiveness of the proposed algorithm for detecting melanoma.