Infrared detectors can be applied to real-time detection systems. They are not affected by weather and light, and they are widely used in infrared guidance for remote sensing [
1]. Infrared imaging is currently being used in many scenarios, including military and civil applications. The technology has promoted the development of small- and medium-sized target recognition and detection in infrared images. Tracking and detecting small targets in space has become the focus of infrared image processing. Infrared image noise originates from the detector, optical components, and circuit systems. At the same time, complex and variable sky backgrounds have an impact on infrared imaging. They are susceptible to interference by clutter and sunlight [
2]. Infrared images lack clear texture and details. Also, the image contrast is poor. It is not easy to recognize small targets. Traditional spatial domain detection methods are suitable for specific imaging environments. They rely on filter templates, resulting in poor robustness for detecting infrared targets in complex backgrounds [
3]. It is difficult to develop a unified framework for small target recognition and detection. Infrared small-target detection utilizes the image’s inherent attributes to enhance the target and suppress the background to achieve the real target [
4].
Traditional infrared small-target detection mainly aims at detecting small cooperative targets. When the signal-to-clutter ratio (SCR) is high, detection before tracking is adopted. When the SCR is low, tracking before detection is applied. Traditional small target detection algorithms include spatial filtering, frequency domain filtering, local contrast-based detection, low-rank, and sparse representation. Spatial filtering contains max–median filtering, max–mean filtering, Gaussian filtering, and morphological filtering methods, etc. Frequency domain filtering mainly includes wavelet transform methods and low-pass filtering algorithms, etc. [
5]. Spatial filtering has high timeliness and has been widely used in engineering applications. Max–median, max–mean, and morphological filtering can achieve highly reliable target detection with specific targets under specified backgrounds. But they are highly dependent on the template size. They show poor robustness to noncooperative small target detection. And they are sensitive to environmental changes [
6].
In local contrast algorithms, researchers have proposed an algorithm based on local contrast measurement (LCM) [
7], which can improve the small target detection rate. But it causes the excessive enhancement of high-brightness point noise, resulting in a high false alarm rate. To solve the shortcomings of LCM, researchers have proposed an improved local contrast measurement algorithm (ILCM). It has resolved the problem of the high false alarm rates. However, the robustness is low due to the excessive reliance on the sliding window [
8]. The article [
9] proposed a new LCM algorithm using a DOG filter (NLCM), which was more robust in complex backgrounds with high noise. However, the method needed to preprocess before the detection and could not adapt to the variable background. The author provided an infrared target detection algorithm based on a high boost-based multiscale local contrast measure (HB-MLCM). It improved the detection effects without complete prior knowledge. Unfortunately, it was time-consuming [
10]. Another article proposed a multiscale relative local contrast (RLCM) [
11]. It could enhance real IR targets and suppress all types of interferences simultaneously. But the imprecise approximation led to false alarms and missed detection. A further article proposed a small-target detection algorithm based on multiscale patch contrast measurement (MPCM) [
12]. Without prior knowledge, the method improved the performance of small target detection. But it could not adaptively select the segmentation thresholds, resulting in a high false alarm rate. A small target detection algorithm using a human visual contrast mechanism and Gaussian difference filter to obtain saliency maps was proposed in [
13]. It improved the detection efficiency; yet, it was not suitable for high-noise environments. The author in [
14] proposed an infrared weak small target detection algorithm (LHM) that calculated local uniformity and local nonuniformity under different scale templates, which had better robustness for noncooperative small target detection. But the computational complexity was higher. The article [
15] proposed an infrared weak and small-target detection algorithm based on multidirectional disparity, which could significantly improve the SCR. But it did not achieve the balance of enhancing the target and suppressing the highlight background edges. Another article proposed an infrared weak and small-target detection method based on the saliency scale space [
16]. They combined saliency images with local contrast differences, which was suitable for scenarios with complex backgrounds. A weighted local contrast enhancement algorithm was proposed in [
17]. They improved the SCRG and suppressed the background clutter at the same time. However, the computational complexity was high, making it unsuitable for engineering applications. At present, the algorithms of multiscale fusion to detect small targets are widely used in rapid and high-performance small-target detection in infrared images [
18,
19,
20]. These methods have achieved better detection results in specific background environments.
Based on the above research, this paper proposes a fused multiscale local gradient contrast algorithm for detecting real infrared small targets in complex backgrounds. The background includes large areas of high-brightness clouds and ground buildings. The infrared small-target detection under different complex backgrounds uses different fusion strategies. We utilize template filter and gradient contrast to finish the multiplicative fusion with a low SCR. When the SCR was higher, we would first adopt template filtering to attain the saliency map. Then, we utilize the new gradient uniformity to calculate the contrast. The method can suppress the real complex noise and highlight the background edge, which can improve the SCR of the target, the detection accuracy, and the efficiency. Compared with the existing high-performance algorithms, the computational complexity is lower. In addition, it is more adaptable to target detection under different complexity backgrounds, which could be used for engineering applications.