sheneman/CerealPestAID-dataset

CerealPestAID Dataset

Image dataset for classifying 26 species of cereal crop pests. Images are organized in ImageFolder format with train/val/test splits.

• Project Website: cerealpestaid.net
• Mobile App:
• Models: sheneman/CerealPestAID
• Source Code: github.com/sheneman/CerealPestAID

Dataset Description

This dataset contains labeled images of 26 cereal pest species commonly found in cereal crop agriculture. The images are organized into subdirectories by species name, compatible with torchvision.datasets.ImageFolder.

• Source: Field-collected pest images
• Format: JPEG images organized in ImageFolder structure
• Splits: train, val, test
• Number of classes: 26
• Total images: 47,136

Split	Images
Train	37,699
Validation	7,056
Test	2,381
Total	47,136

Directory Structure

data/
├── train/
│   ├── Cabbage_seedpod_weevil/
│   ├── bird_cherry_oat_aphid/
│   ├── cabbage_aphid/
│   ├── ... (26 class directories)
├── val/
│   ├── Cabbage_seedpod_weevil/
│   ├── bird_cherry_oat_aphid/
│   ├── ... (26 class directories)
└── test/
    ├── Cabbage_seedpod_weevil/
    ├── bird_cherry_oat_aphid/
    ├── ... (26 class directories)

Class Distribution

The dataset exhibits significant class imbalance, with the largest class (Cutworms) containing over 166x more images than the smallest (Turnip aphid).

Index	Species	Train	Val	Test	Total
0	Cabbage seedpod weevil	490	91	32	613
1	Bird cherry oat aphid	517	97	33	647
2	Cabbage aphid	1,538	288	97	1,923
3	Cereal grass aphid	446	83	29	558
4	Cereal leaf beetle	1,068	200	68	1,336
5	Clickbeetles / wireworms	1,644	307	104	2,055
6	Crucifer flea beetle	250	46	17	313
7	Cutworms	7,842	1,470	491	9,803
8	Diamondback moth	6,496	1,218	406	8,120
9	English grain aphid	1,053	197	67	1,317
10	Greenbug	52	9	4	65
11	Green peach aphid	305	57	20	382
12	Hessian fly	106	19	8	133
13	Lygus bug	3,452	647	216	4,315
14	Non-pest herbivores	220	41	15	276
15	Occasional pest	972	182	61	1,215
16	Pea aphid	285	53	19	357
17	Pea leaf weevil	2,492	467	157	3,116
18	Pea weevil	372	69	24	465
19	Predators	1,952	366	123	2,441
20	Rose grain aphid	104	19	7	130
21	Russian wheat aphid	145	27	10	182
22	Stink bug	5,005	938	314	6,257
23	Striped flea beetle	278	52	18	348
24	Turnip aphid	47	8	4	59
25	Wheathead armyworm	568	105	37	710

Handling Class Imbalance

This dataset has substantial class imbalance (e.g., Cutworms has 9,803 images while Turnip aphid has only 59). The associated training code addresses this using PyTorch's WeightedRandomSampler with inverse class frequency weighting. For each sample, a weight is assigned as the inverse of its class count (1 / class_count), ensuring that minority classes are sampled proportionally more often during training:

from torch.utils.data import WeightedRandomSampler
import numpy as np

class_counts = np.bincount([label for _, label in dataset.imgs])
class_weights = 1.0 / class_counts
sample_weights = [class_weights[label] for _, label in dataset.imgs]

sampler = WeightedRandomSampler(sample_weights, num_samples=len(sample_weights), replacement=True)

This approach effectively balances the training distribution so the model sees each class with equal probability per epoch, mitigating bias toward overrepresented classes without discarding any data.

Usage

With torchvision

from torchvision import datasets, transforms

transform = transforms.Compose([
    transforms.Resize(572),
    transforms.CenterCrop(528),
    transforms.ToTensor(),
    transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
])

train_dataset = datasets.ImageFolder(root="train", transform=transform)
val_dataset = datasets.ImageFolder(root="val", transform=transform)
test_dataset = datasets.ImageFolder(root="test", transform=transform)

With HuggingFace datasets

from datasets import load_dataset

dataset = load_dataset("sheneman/CerealPestAID-dataset")

Associated Models

Pre-trained classifiers for this dataset are available at sheneman/CerealPestAID.

Model	Test Accuracy
EfficientNet-B6	92.94%
MobileNetV3-Large	90.09%
InceptionV3	79.00%

Acknowledgments

This project, titled "Harnessing Artificial Intelligence for Implementing Integrated Pest Management in Small-Grain Production Systems," is funded under the U.S. Department of Agriculture No. 2021-67021-34253.

Team

• Sanford Eigenbrode - Distinguished Professor, Entomology, Plant Pathology, and Nematology, University of Idaho (PI)
• Arash Rashed - Virginia Tech Southern Piedmont Agricultural Research and Extension Center
• Marek Borowiec - Assistant Professor, Insect Systematist, Director of C. P. Gillette Museum, Colorado State University
• Subodh Adhikari - Assistant Professor, Entomology Extension Specialist, Utah State University
• Luke Sheneman - Director of Research Computing, University of Idaho
• Jennifer Hinds - Research Applications Architect, University of Idaho
• John Brunsfeld - Senior Full Stack Developer, University of Idaho

Citation

@misc{sheneman2025cerealpestaid,
  author = {Sheneman, Luke and Borowiec, Marek and Eigenbrode, Sanford and Rashed, Arash and Adhikari, Subodh and Hinds, Jennifer and Brunsfeld, John},
  title = {CerealPestAID: Deep Learning Models for Cereal Pest Identification},
  year = {2025},
  publisher = {HuggingFace},
  url = {https://huggingface.co/sheneman/CerealPestAID}
}

License

MIT