When using remote sensing data to study vegetation characteristics, it is often challenging to acquire or collect a large dataset that is accurately labeled. This difficulty is usually due to several issues, including but not limited to the study site's spatial area and accessibility, errors in the global positioning system, and mixed pixels caused by an image's spatial resolution. We propose an approach that estimates multiple target signatures from training samples with imprecise labels: Multi-Target Multiple Instance Adaptive Cosine Estimator (Multi-Target MI-ACE). The proposed method addresses the problems above by directly considering the multiple-instance, imprecisely labeled dataset, which eliminates the need for a pixel by pixel labeled dataset. The method learns a dictionary of target signatures that optimizes detection against a background using the Adaptive Cosine Estimator (ACE). We used this algorithm to detect the location of multiple plant functional types (PFTs) using an AVIRIS hyperspectral dataset collected over Santa Barbara County, California. Overall, PFTs were detected with high accuracy even though the training dataset was mixed. PFTs that are spectrally more homogenous due to the plant species sharing similar plant properties performed the best. PFTs with more spectral variability due to significant differences in species or open canopies had lower performance. The Multi-Target MI-ACE algorithm was able to handle all spectral variability that is inherent in all plant datasets, which is caused by differences in properties such as plant structure, biochemistry, and water status. The Multi-Target MI-ACE algorithm proved effective for determining PFTs while also be able to handle mixed and ambiguously labeled training data.