Monitoring Endangered and Rare Wildlife in the Field: A Foundation Deep Learning Model Integrating Human Knowledge for Incremental Recognition with Few Data and Low Cost

Abstract

Intelligent monitoring of endangered and rare wildlife is important for biodiversity conservation. In practical monitoring, few animal data are available to train recognition algorithms. The system must, therefore, achieve high accuracy with limited resources. Simultaneously, zoologists expect the system to be able to discover unknown species to make significant discoveries. To date, none of the current algorithms have these abilities. Therefore, this paper proposed a KI-CLIP method. Firstly, by first introducing CLIP, a foundation deep learning model that has not yet been applied in animal fields, the powerful recognition capability with few training resources is exploited with an additional shallow network. Secondly, inspired by the single-image recognition abilities of zoologists, we incorporate easily accessible expert description texts to improve performance with few samples. Finally, a simple incremental learning module is designed to detect unknown species. We conducted extensive comparative experiments, ablation experiments, and case studies on 12 datasets containing real data. The results validate the effectiveness of KI-CLIP, which can be trained on multiple real scenarios in seconds, achieving in our study over 90% recognition accuracy with only 8 training samples, and over 97% with 16 training samples. In conclusion, KI-CLIP is suitable for practical animal monitoring.

Utilizing Geographical Distribution Statistical Data to Improve Zero-Shot Species Recognition

“…Among them, prompt engineering was the key to improving zero-shot recognition performance. To design appropriate prompts to improve the performance of zero-shot and few-shot species recognition, both Maniparambil et al [ 22 ] and Mou et al [ 39 ] designed prompts using species’ visual descriptions to improve the performance of few-shot species recognition. Parashar et al [ 24 ] analyzed the dataset for training CLIP and found that the scientific name of the species appeared less often than the common name, and therefore designed prompts by converting the scientific name of the species to the common name.…”

Artificial intelligence and machine learning applications in forest management and biodiversity conservation

“…The subsequent sections delineate the economic markets for AI applications in the forestry sector and the resultant proliferation of promising AI technology start-ups and nonprofit organizations worldwide. These entities strive to digitize forests, enhance forest management, mitigate escalating CO2 levels, safeguard endangered animal species, combat wildlife trafficking and illegal trading, facilitate wildlife census alongside monitoring, and automate taxonomic recognition and classification of plants and animals by embracing digital advancements [29][30][31][32][33][34][35][36][37][38][39][40][41] .…”

Improved Wildlife Recognition through Fusing Camera Trap Images and Temporal Metadata

“…de Lutio et al [16] utilized the spatial, temporal, and ecological contexts attached to most plant species' observation information to construct a digital taxonomist that improved accuracy from 73.48% to 79.12% compared to a model trained using only images. Mou et al [17] used animals' visual features, for example, the color of a bird's feathers or the color of an animal's fur, to improve the recognition accuracy of a contrastive language-image pre-trained (CLIP) model on multiple animal datasets. Camera traps in national parks are capable of monitoring wildlife continuously for long periods of time with reduced human intervention and can provide complete information on animal rhythms.…”