Detection Seminar Resources

Kynology presents “Next Level Detection”

ATK9 Conference 2025

Instructors: Angel Landrau, DJ Holmes and Dr. Stewart Hilliard

Summary of lecture on detector K9 theory

Part 1—Alert vs Indication

  • The detector dog’s response to target (trained) odor occurs in two phases, the alert–also called change of behavior– and then the indication– also called final response
  • Here we will use the law enforcement terminology—alert and indication. When referring to both alert and indication together, we will use “detection responding.”
  • The alert to trained odor is based on the Pavlovian odor – reward association (i.e. nitro dynamite odor – rubber ball)
  • The indication of trained odor is based on the instrumental response – reward association (i.e. sit – rubber ball)
  • Some authorities consider the alert more important (i.e. reliable) than the indication, based on the idea that the Pavlovian “alert” is involuntary, more or less a Pavlovian conditioned reflex, and therefore the dog cannot be mistaken or “lie” with respect to the alert
  • Some authorities consider the indication more important (i.e. reliable) based on the idea that the indication is an objective and easily-described behavior that is “built” through reward history so that it occurs only in response to target odor
  • When considering the relative importance of the alert vs the indication, every handler should know the following about their dog:
  • Is it possible to tell the difference between the dog’s alert response to trained odor and the dog’s “interest” response when investigating a non-target odor? If this is not possible then the handler may be mistaken when “calling” a find based on the alert
  • What is the dog’s probability of an indication response to a non-target odors (i.e. a false response or FR to a non-target)? If the dog’s FR probability is significantly greater than zero on non-targets, then the handler may be mistaken when “calling” a find based on the indication
  • Detector dogs should be repeatedly exposed to a wide variety of non-target odors so that:
  • The handler has abundant opportunity to compare the dog’s “alert” to target odor and “interest” in non-targets, and attempt to distinguish between the two
  • The dog receives extensive training in discriminating target odor from non-targets

Part 2—Reward schedules

  • In order to teach a dog a new skill, it is best to use a continuous reinforcement schedule, meaning that the dog is rewarded every time that it correctly responds
  • In order to make the skill persistent (long-lasting, durable, stubborn), it is best to transition to random reinforcement schedules, meaning that the dog is not rewarded for every correct response but instead for only some of those responses, in a random fashion
  • Detector dogs must be very persistent while searching, and it is normal to reward a dog randomly for search behavior. One can think of this as rewarding the dog for varible numbers of “sniffs,” or rewarding the dog for variable periods of search. This is commonly-accepted practice.
  • It is not accepted practice to reward detector dogs in a partial random manner for the indication. Most handlers reward the dog for every correct final response (i.e. use continuous reinforcement) in training
  • However, during operations there are occasions when it is advisable to withhold reward for an indication (I.e. in the case of an explosives detector dog that gives an indication in a suspected laboratory for manufacture of unstable home-made explosive such as TATP)
  • It is therefore best practice to introduce the dog to random reward for indication during training, so that the dog does not interpret being “pulled off” of a find as punishment for a mistake

Part 3—Signal detection theory

  • Detector dogs are essentially signal receivers tuned for detecting an informative signal (the odor of a target substance) against a background of other, non-target odors
  • How accurate the dog is at detecting the odor signal in training or certification evaluations is measured using two ratios, or fractions:
  • Number of correct detections (“Hits) over the total number of targets available
  • Number of false indications (“FRs”) over some arbitrary number—Very often the number used is 1, standing for the certification or the day’s training
  • Thus it is common in the detector dog industry to describe a standard for acceptable detector dog performance in terms such as 90% or better Hits (9 Hits/10 available targets), and 2 or less FRs per certification (2 FRs/1= 2)
  • This way of describing detector dog performance neglects the inverse of FR’s—correct rejections—It does not directly express how often the dog did not show detection responding when target odor was not present, in effect correctly saying “no, the target is not here.”
  • In order to be reliable and accurate the detector dog must actually correctly say “no, my target odor is not present” far more often than it must correctly say “yes, my target odor is present,” yet in conventional training we do not reward correct rejections. We reward only correct dteections, or hits
  • The fact that reward is available only for saying “yes” biases the dog towards saying “yes,” creates a psychological condition (called a criterion) that makes it difficult to reduce the dog’s FR rate to zero or near zero
  • Trainers tend to concentrate on FRs. They focus on how to reduce FRs through some form of punishment, most commonly by withholding reward, but sometimes also through physical punishment
  • An alternate and promising approach for reducing the dog’s FR rate is to focus on correct rejections and reward them. Providing rewards for correct rejections reduces the dog’s bias towards saying “yes,” and thus the dog’s rate of FRs.
  • When we reward a dog for continued searching past a particular search area (a cabinet, say) we are in effect rewarding for correctly rejecting the cabinet, for saying “No, that cabinet did not smell of target odor.”
  • Next Level instructors teach specific protocols for rewarding search/correct rejection.

Part 4—Generalization and discrimination

  • Stimulus control is the property of a cue or command enabling it to “trigger” or elicit a given trained response
  • In detector dog training the cue we are most concerned with is target odor– which is intended to very reliably elicits or “controls” the dog’s indication response
  • Generalization is a description of how specific, or “strong,” is target odor’s control of the indication response.
  • When generalization is very broad, detection responding is elicited not only by target odor, but by a wide range of other odors. Broad generalization is equal to weak stimulus control by target odor
  • When generalization is very narrow, detection responding is elicited only by target odor, and not by any other odors. Narrow generalization is equal to strong stimulus control by target odor
  • To make control of the dog’s detection response strong, to restrict generalization and make the dog’s indication specific to target odor, we perform discrimination training, in which we reward the dog for responding to target odor but not for responding to a wide range of other odors

Part 5—Stimulus control and targets

  • Strong stimulus control by target odor means that target odor is both necessary and also sufficient to trigger detection responding
  • Necessary = target odor must be present for detection response to occur
  • Sufficient = nothing besides target odor is needed for the dog to respond
  • However, because operational targets virtually never smell exactly like training targets, stimulus control of the dog’s detector response can be too strong, too specific. When (training) target odor stimulus control is too strong, the dog fails to generalize from its’ trained odor to the odor of operational targets (or perhaps the targets used in a certification in the next county over)
  • If stimulus control is too weak and generalization too broad, then the detector dog will respond not only to training targets and operational targets, but also false respond on non-target substances
  • For accuracy the detector dog must have a properly “tuned” generalization curve, so that it generalizes from its’ trained target odors to similar but not identical target odors encountered in operations, but does not generalize to non-target substances
  • We can summarize procedures to properly tune the dog’s generalization curve as follows:
    n Ensure the dog’s responding is under control of training target odor, and that non-targets (odors and other stimuli) have minimal stimulus control over responding
  • n Train using all available variants of the target substance from early in training
  • n Expose the dog to all possible non-targets from very early in training, to maximize the dog’s discrimination ability, and also weaken any influence by non-target odors and other stimuli
  • n Later in training, and on operations, prioritize training with target substances recovered from the area of operations

Part 6—Stimulus control and non-targets

  • A very wide range of non-target stimuli—odor, visual, auditory, from humans, from animals—can interfere with accuracy by acting in one of two ways: as distractors or as associated stimuli
  • Distractors are stimuli that do not have any reliable association with target or with reward. They are simply noticeable stimuli in the environment that capture the dog’s attention and interfere with searching for and detecting target odor
  • Distractors can bias the dog away from correct alert/indication (causing the dog to miss) simply by disrupting the dog’s search focus, or by “masking” target odors.
  • Distracters can bias the dog towards responding incorrectly (FRs) through “surprise” or novelty” effects, or generalization along dimensions like “smells sort of ‘chemically,’” or “smells rotten,” or “doesn’t seem to belong here, it is not like the background.”
  • Associated stimuli are stimuli that are likely to become linked with target odor or with reward by appearing with target odor and/or appearing immediately before reward. As a result they are at very high risk to develop stimulus control over the dog’s detection response because the dog associates them with target or because the dog associates them with reward
  • Classic examples of associated stimuli are the visual and odor cues of odor delivery devices such as scent boxes, canvas bags, plastic or glass tubes, aluminum boxes, etc.
  • Associated odors may bias the dog away from correct responding/hits when they are absent because through association they have developed enough stimulus control to support the dog’s detection response. “Support” means that when associated odors are present they “confirm” the dog’s response.
  • When confirming associated stimuli are absent, the probability of the dog responding correctly may be decreased. When confirming associated stimuli are present in training but not present on operations, then operational effectiveness is likely to be decreased
  • Associated odors may bias the dog towards incorrect responding/FRs because they have developed enough stimulus control to trigger false responses in isolation, or when multiple associated stimuli are present and their accumulated stimulus control is sufficient to trigger a FR
  • Therefore it is necessary to perform training in such a fashion that associated odors develop no more than minimal stimulus control over responding–

Part 7—Minimizing stimulus control of associated stimuli

  • Minimization of stimulus control by associated stimuli depends upon weakening the relationship between associated stimuli and targets, and weakening the relationship between associated stimuli and reward
  • In order to understand minimizing stimulus control of associated odors, it is necessary to understand some basic rules about how stimuli learning about various stimuli is affected when those stimuli are not presented one at a time before reward, but presented simultaneously before reward, or in compound.
  • Stimuli presented in compound compete with each other for stimulus control. Three main processes determine which stimulus wins the competition and becomes associated
  • Blocking takes place when a stimulus that is already associated with reward prevents conditioning of a second stimulus. For example, if a dog has already learned that tin can delivery device odor predicts reward, and then tin can odor and methamphetamine are presented together (methamphetamine inside the tin can) prior to reward, the dog may fail to learn that amphetamine odor predicts reward. In blocking a previously-conditioned odor blocks conditioning of a new odor when they are presented in compound.
  • Overshadowing takes place when a stimulus that is new but very intense or noticeable (i.e. salient) prevents conditioning of another new stimulus that smells much weaker. For e.g. if a dog has never smelled copper pipe (a strong odor) or TNT (a weaker odor) and we attempt to train recognition of TNT by presenting a small amount of TNT in a copper pipe, the dog may fail to learn TNT odor because the intense copper odor overshadows TNT. In overshadowing an intense new odor prevents conditioning/imprinting of a weak new odor when they are presented in compound.
  • There is a third process in which one stimulus can diminish the stimulus control of other stimuli presented in compound that depends neither on previous conditioning or on differences in salience. This process is based on relative validity, or which stimulus is a better predictor of reward.
  • Relative validity is especially important where odor delivery devices are concerned because inevitably nearly all target substances will be presented in compound with the odor of some container or enclosure .
  • If a dog experiences reward after 100% of encounters with TNT odor, but only after 33% of encounters with plastic tube odor, then TNT has much better reward prediction validity relative to plastic tube odor, and TNT will reduce the stimulus control developed by plastic tube odor
  • Detector dog trainers can manipulate the predictive value (or validity) of odor delivery device odors in two main ways:
  • Provide clean/empty odor delivery devices for the dog to encounter while searching as often as possible
  • Use multiple odor delivery device materials to present target odor whenever possible (e.g. smokeless powder in plastic tube, metal tube, and cardboard tube)