Over millions of years, mammals developed sophisticated biological neural systems to recognize life-threatening insects. Humans have taken this ability to the next level by developing cognitive-cultural systems to classify them in a more rigorous way. From recognizing insects that are poisonous, to recognizing insects that damage our crops and spread diseases, to recognizing them for building a taxonomic map of their biological diversity.
This is not enough. Our creative nature demands that we go up to the next level: automatically recognize insects through electronic devices in order to build a system to monitor their movement geographically at a global level as hurricanes are currently monitored. Also, to assist humans who do not have the experience to recognize them and free the experts from this tedious task. Computerizing this task seems easy at first sight, but it presents a list of technical challenges that we study in this short report. We divide these challenges into two large groups: 1) intrinsic challenges: they are due to the general properties of insects, and 2) extrinsic challenges: they are due to the technology we use to recognize them, including hardware and software.
Intrinsic challenges
Granularity in the taxonomy. The taxonomy of insects has several hierarchical levels: from kingdom to species through class, order and family such as the taxonomy of a bedbug shown in Fig 1. This allows the same insect to be recognized at different levels of taxonomic generalization, adding greater difficulty to the insect recognition. For pest control purposes, this difficulty is simplified as only the family or genus is recognized, ignoring the lower levels. For finer purposes, this problem must be solved with greater granularity in the taxonomy.
Unknown number of insect classes. As if that were not enough to recognize the taxonomy of documented insects, it turns out that this taxonomy is not complete because we know only 20% of all the insects on the planet [1] (including the rest of arthropods). This is a major problem scientifically, but it is smoothed out for pest control applications since new insect pests do not appear often.
Appearance variability. It is more difficult to recognize the same insect when it changes its appearance. Their appearance can change for various reasons: their life cycle, gender differences, and other conditions such as whether the insect has recently eaten or not. In Fig. 2, we illustrate how the appearance of bedbugs and ladybugs changes through their life cycle. In the case of bedbugs also due to their state of nutrition. It is clear that the appearance of ladybugs can be very different depending on their stage of life. In contrast, bedbugs have more stability in their appearance.
In the next post, we will continue to explain the extrinsic challenges for recognizing insects automatically.
References
[1] Stork, N. E. (2018). How many species of insects and other terrestrial arthropods are there on Earth?. Annual review of entomology, 63, 31-45.