Infrared target detection is a challenging computer vision problem which involves detecting small targets in heavily cluttered conditions while maintaining a low false alarm rate. We propose a network that optimizes a “target to clutter ratio”(TCR) metric defined as the ratio of the output energies produced by the network in response to targets and clutter. A TCR-network (TCRNet) is presented in which the filters of the first convolutional layer are composed of the eigenvectors most responsive to targets or to clutter. These vectors are analytically derived via a closed form optimization of the TCR metric. The remaining convolutional layers are trained using a novel cost function also designed to optimize the TCR criterion. We evaluate the performance of the TCRNet using a public domain medium wave infrared dataset released by the US Army's Night Vision Laboratories, and compare it to the state-of-the-art detectors such as Faster regions with convolutional neural networks (R-CNN) and Yolo-v3. The TCRNet demonstrates state-of-the-art results with greater than 30% improvement in probability of detection while reducing the false alarm rate by more than a factor of two when compared to these leading methods. Experimental results are shown for both day and night time images, and ablation studies are presented which demonstrate the contribution of the first layer eigenfilters, additional convolutional layers, and the benefit of the TCR cost function.

中文翻译：

---

通过使用卷积神经网络最大化目标杂波比（TCR）指标，在杂乱环境中进行红外目标检测

红外目标检测是具有挑战性的计算机视觉问题，涉及在严重混乱的情况下检测小目标，同时保持较低的误报率。我们提出了一种优化“目标杂波比”（TCR）指标的网络，该指标定义为网络响应目标和杂波而产生的输出能量之比。提出了一种TCR网络（TCRNet），其中第一卷积层的滤波器由对目标或杂波最敏感的特征向量组成。这些向量是通过TCR度量的封闭形式优化分析得出的。剩余的卷积层使用新颖的成本函数进行训练，该函数也设计为优化TCR标准。我们使用美国陆军发布的公共领域中波红外数据集评估TCRNet的性能。s夜视实验室，并将其与最新的探测器进行比较，例如带卷积神经网络（R-CNN）和Yolo-v3的Faster区域。与这些领先方法相比，TCRNet展示了最新结果，检测概率提高了30％以上，同时将误报率降低了两倍以上。显示了白天和晚上的图像的实验结果，并进行了消融研究，这些研究表明了第一层特征滤波器，其他卷积层的贡献以及TCR成本函数的优势。与这些领先方法相比，TCRNet展示了最新结果，检测概率提高了30％以上，同时将误报率降低了两倍以上。显示了白天和晚上的图像的实验结果，并进行了消融研究，这些研究表明了第一层特征滤波器，其他卷积层的贡献以及TCR成本函数的优势。与这些领先方法相比，TCRNet展示了最新结果，检测概率提高了30％以上，同时将误报率降低了两倍以上。显示了白天和晚上的图像的实验结果，并进行了消融研究，这些研究表明了第一层特征滤波器，其他卷积层的贡献以及TCR成本函数的优势。