The effect of training paradigm on dogs’ (Canis familiaris) acquisition and generalization of smokeless powders

Abstract

Explosives detection canines need to be trained to detect a range of energetic variations at an efficient scale to meet the needs of public safety. The goal of this study was to determine which of three training paradigms most efficiently trained canines to detect eight smokeless powder (SP) varieties and to generalize to novel SP exemplars. Three training paradigms were compared: Sequential (dogs were trained to one odor at a time), Mixture (dogs were trained to a mixture of four SPs), and Inter-mixed (dogs were trained to four unmixed SPs within a session). SPME GC-MS was used to evaluate the volatile organic compound (VOC) profile of 18 SPs. We then identified four SPs that showed maximal VOC profile variation as the initial training targets and four related variants. The training paradigm had no effect on acquisition time. One anomalous SP was observed, where 12 of 18 dogs failed to reach criterion. Inter-mixed training led to higher rates of generalization across 10 untrained SP varieties compared to Sequential and Mixture training. Mixture and Sequential trained dogs did not differ in their generalization rates. Although Inter-mixed training led to higher rates of generalization, it did not produce proficiency in detection of all novel variants, with many variants (double and single base) producing < 75 % response rates. Explicit training with some variants is still required with Inter-mixed training. A Partial Least Squares Regression (PLS), where the SPME GC/MS VOC peak areas for each detected VOC were used to predict canine nose hold time, explained 72 % of the variance in VOC data and 17 % of the variance in cumulative nose hold time. The VOCs with the highest variable importance was one VOC unique to Vihta Vuori® single bases, and two VOCs associated with Hodgdon® single bases, to which dogs showed poor generalization. Good to adequate generalization was observed across Hodgdon® double bases and Accurate® single and double bases. Overall, these results suggest that Inter-mixed training has important benefits in generalization; however, explicit training is needed to reach proficiency.