An approach to rapid processing of camera trap images with minimal human input

Abstract

Camera traps have become an extensively utilized tool in ecological research, but the manual processing of images created by a network of camera traps rapidly becomes an overwhelming task, even for small camera trap studies. We used transfer learning to create convolutional neural network (CNN) models for identification and classification. By utilizing a small dataset with an average of 275 labeled images per species class, the model was able to distinguish between species and remove false triggers. We trained the model to detect 17 object classes with individual species identification, reaching an accuracy up to 92% and an average F1 score of 85%. Previous studies have suggested the need for thousands of images of each object class to reach results comparable to those achieved by human observers; however, we show that such accuracy can be achieved with fewer images. With transfer learning and an ongoing camera trap study, a deep learning model can be successfully created by a small camera trap study. A generalizable model produced from an unbalanced class set can be utilized to extract trap events that can later be confirmed by human processors.

Trade-off between deep learning for species identification and inference about predator-prey co-occurrence: Reproducible R workflow integrating models in computer vision and ecological statistics

“…We calculated three metrics to evaluate our model performance at correctly identifying species (e.g. Duggan et al 2021). Specifically, we relied on accuracy the ratio of correct predictions to the total number of predictions, recall a measure of false negatives (FN; e.g.…”

“…an image with any species but a lynx for which our model predicts a lynx) with precision = TP / (TP + FP). In camera trap studies, a strategy (Duggan et al 2021) consists in optimizing precision if the focus is on rare species (lynx), while recall should be optimized if the focus is on commom species (chamois and roe deer).…”

Using informative priors to account for identifiability issues in occupancy models with identification errors

“…Within the context of sensor-based assessment method, data are massive and needs to be processed before being analyzed. This consists in identifying the taxon of interest in a large amount of collected data, referred to as identification or classification, which can be carried out either manually by one or several operators (Swanson et al, 2015; Welbourne et al, 2015), either automated using deep learning algorithms (Duggan et al, 2021; Tabak et al, 2019), or a combination of both (Augustine et al, 2023; Campos-Cerqueira and Aide, 2016). For images and acoustic data, combining manually and automatically-generated data processing helps to control classification errors; when one species is mistaken for another (Barré et al, 2019).…”

The successes and pitfalls: Deep‐learning effectiveness in a Chernobyl field camera trap application

“…Due to the considerable time and effort expended by researchers when classifying camera trap images, many studies have deployed the use of machine learning to rapidly classify animal species and anthropogenic objects, including humans and vehicles (Duggan et al, 2021 ; Tabak et al, 2018 ). In fact, some studies have even found that machine‐learning models can sometimes outperform the average citizen scientist with regard to accuracy (Norouzzadeh et al, 2018 ; Whytock et al, 2021 ).…”