<p>Infrared maritime imaging has a wide application of distress target salvage and rescue. The detection of small targets is the most valuable research in this field. Although existing CNN-based networks have achieved a promising result for ordinary targets, infrared maritime small targets are still far from satisfactory because of their small size and little detail information. As network layers deepen and the number of downsampling operations increases, infrared small target presents a single feature in the feature maps, causing missing detection of small targets. This paper proposes a novel infrared maritime target detection network, CFSLFE, incorporating cascade fusion and a spatial local feature enhancement module to address these challenges. First, we design an appropriate IoU threshold and propose a new weight-modulated loss function tailored for small targets that are particularly sensitive to IoU. Second, cascade fusion, which utilizes residual network and dilated convolution, is designed to integrate the features into backbone under different scales and allow more parameters to be used for feature fusion. Third, an effective and simple attention module is proposed for the feedforward convolutional neural network, which is called a spatial local feature enhancement module. This module is designed to address the single feature of small targets in feature maps before transmitting to small target detection head, which enhances the local saliency of small targets while suppressing a large number of background clutter with similar structure within the local region. The results show that CFSLFE outperforms state-of-the-art methods on the infrared maritime target dataset. The code and IRSTD-1K dataset are available at <a href="https://github.com/kuanshi1718/CFSLFE">https://github.com/kuanshi1718/CFSLFE</a>; <a href="https://github.com/RuiZhang97/ISNet">https://github.com/RuiZhang97/ISNet</a>.</p>