2024/12/27 by Xu, Wenxiu, Saba Ghorbani Bazegar, Dong Sheng +8
Agricultural and Biological Sciences · #Artificial Intelligence (cs.AI) #Cocoa and Sweet Potato Agronomy #FOS: Biological sciences #FOS: Computer and information sciences #Food Chemistry and Fat Analysis #Plant and animal studies #Quantitative Methods (q-bio.QM)
paper · pdf · doi:10.48550/arxiv.2412.19915
openalex publication_date 2024/12/27 · openalex created_date 2025/01/01 · openalex updated_date 2026/07/28
1. Description of the cocoaFlowerVisitorsDataset We present the first cocoa flower visitor dataset consisting of 5,792 insect images and 1,082 flower ‘background’ images. Of these, 5,214 insect images and 782 background images were collected in 2023, featuring five common cocoa flower visitors: Ceratopogonidae (midges), Formicidae (ants), Aphididae (aphids), Araneae (spiders), and Encyrtidae (parasitoid wasps). In 2024, 578 insect images and 300 background images were collected, featuring three common cocoa flower visitors: Ceratopogonidae, Formicidae, and Encyrtidae. The dataset was curated from 23 million images collected over two years by embedded cameras20 deployed in cocoa plantations at the Xinglong Tropical Botanical Garden, Hainan Province, China. Details are provided in the preprint entitled “Identifying Cocoa Pollinators: A Deep Learning Dataset” by Wenxiu Xu, Saba Ghorbani Bazegar, Dong Sheng, Manuel Toledo-Hernández, ZhenZhong Lan, and Thomas Cherico Wanger. Description of the cocoaFlowerVisitorsDataset file structure The zip file contains four folders: background images » background cocoa flower images of the 2023 & 2024 field season test (2024) » test images and labels of five classes from the 2024 field season train (2023) » test images and labels of five classes from the 2023 field season val (2023) » validation images and labels of five classes from the 2023 field season 2. Description of the cocoaFlowerVisitorsCode The code used in this study is based on the open-source Ultralytics YOLO repository (https://github.com/ultralytics/ultralytics). Only minor parameter modifications (e.g., training epochs, image size, confidence thresholds) were made to adapt the model to our dataset. Here are all the scripts used to run training and inference, along with the configuration files.