A comparative study of target detection algorithms in hyperspectral imagery applied to agricultural crops in Colombia

Background: (HSI) Hyperspectral Images contain high spectral resolution information, in hundreds of contiguous bands over a specific range of the electromagnetic spectrum. In science and industry, hyperspectral information is exploited by means of classification, anomaly and target detections algorithms. Specifically, in the last two decades a wide variety of hyperspectral target detection algorithms have been proposed. However, an optimal target detection algorithm with a remarkable performance over different kinds of targets and scenarios is still an active matter of research, due to the high spectral variability and diversity of real-world scenarios. Aim: This work presents a comparative study of target detection algorithms in hyperspectral imagery applied to agricultural crops in Colombia for evaluate performance in different scenarios. Method: The evaluations were performed on 20 real HSI acquired by the satellite Hyperion sensor, and 6 synthetic HSI with different noise levels. 5 synthetic targets were implemented; more than 115 spectral real signatures were extracted, 11 of those signatures were used as target in the testing process, allowing to characterize 5 agricultural crops of Colombian northeastern in 5 different areas. Results: The results show that the Adaptive Coherence Estimator (ACE) algorithm has a better performance in terms of detection probabilities PD > 90% for different scenarios and targets of agricultural type, in both synthetic and real images. Conclusions: In applications for target detection in HSI, it is critical to find an algorithm to have optimal performance for different scenarios and targets, due to the spectral variability generated by the geographical conditions countrywide. On the other hand, this work shows that is possible the development of new research fields and applications at the national level, taking advantage of hyperspectral imaging techniques for spectral detection, specifically for Colombian agriculture