In remote sensing, it is often difficult to acquire or collect a large dataset that is accurately labeled. This difficulty is often due to several issues including but not limited to the study site's spatial area and accessibility, errors in global positioning system (GPS), and mixed pixels caused by an image's spatial resolution. An approach, with two variations, is proposed that estimates multiple target signatures from mixed training samples with imprecise labels: Multi-Target Multiple Instance Adaptive Cosine Estimator (Multi-Target MI-ACE) and Multi-Target Multiple Instance Spectral Match Filter (Multi-Target MI-SMF). The proposed methods address the problems above by directly considering the multiple-instance, imprecisely labeled dataset and learns a dictionary of target signatures that optimizes detection using the Adaptive Cosine Estimator (ACE) and Spectral Match Filter (SMF) against a background. The algorithms have two primary steps, initialization and optimization. The initialization process determines diverse target representatives, while the optimization process simultaneously updates the target representatives to maximize detection while learning the number of optimal signatures to describe the target class. Three designed experiments were done to test the proposed algorithms: a simulated hyperspectral dataset, the MUUFL Gulfport hyperspectral dataset collected over the University of Southern Mississippi-Gulfpark Campus, and the AVIRIS hyperspectral dataset collected over Santa Barbara County, California. Both simulated and real hyperspectral target detection experiments show the proposed algorithms are effective at learning target signatures and performing target detection.