COTONET: A custom cotton detection algorithm based on YOLO11 for stage of growth cotton boll detection

Guillem González; Guillem Alenyà; Sergi Foix

cs.CVMar 12, 2026

COTONET: A custom cotton detection algorithm based on YOLO11 for stage of growth cotton boll detection

Authors: Guillem González, Guillem Alenyà, Sergi Foix

Organizations: Institut de Robòtica i Informàtica Industrial (CSIC-UPC), Llorens i Artigas 4-6, Barcelona, 08028, Spain

Abstract

Cotton harvesting is a critical phase where cotton capsules are physically manipulated and can lead to fibre degradation. To maintain the highest quality, harvesting methods must emulate delicate manual grasping, to preserve cotton's intrinsic properties. Automating this process requires systems capable of recognising cotton capsules across various phenological stages. To address this challenge, we propose COTONET, an enhanced custom YOLO11 model tailored with attention mechanisms to improve the detection of difficult instances. The architecture incorporates gradients in non-learnable operations to enhance shape and feature extraction. Key architectural modifications include: the replacement of convolutional blocks with Squeeze-and-Exitation blocks, a redesigned backbone integrating attention mechanisms, and the substitution of standard upsampling operations for Content Aware Reassembly of Features (CARAFE). Additionally, we integrate Simple Attention Modules (SimAM) for primary feature aggregation and Parallel Hybrid Attention Mechanisms (PHAM) for channel-wise, spatial-wise and coordinate-wise attention in the downward neck path. This configuration offers increased flexibility and robustness for interpreting the complexity of cotton crop growth. COTONET aligns with small-to-medium YOLO models utilizing 7.6M parameters and 27.8 GFLOPS, making it suitable for low-resource edge computing and mobile robotics. COTONET outperforms the standard YOLO baselines, achieving a mAP50 of 81.1% and a mAP50-95 of 60.6%.

Figures & tables

Figure 1: Two types of leaves found in cotton plants. On the left, a sample of a bract. On the right, a sample of a true leaf.

Panel of Figure 2 — Figure 2: Different types of images found within the dataset: (a) Clear, close view (b) Medium-far view (c) Sunny (d) Cloudy (e) Leaf occlusion (f) Cotton occlusion (g) Blur by focal distance (h) Perpendicular view (i) Angled view.

Figure 3: The Squeeze and Excitation (SE) module.

Figure 4: The simple attention (SimAM) module.

Figure 5: The channel gate from the PHAM block.

Figure 6: The spatial gate from the PHAM block.

Figure 7: The coordinate gate from the PHAM block.

Figure 8: The Parallel Hybrid Attention Module (PHAM).

Figure 9: The Content Aware ReAssembly of FEatures (CARAFE) module.

Figure 10: COTONET, the proposed YOLO11 architecture modification.

CPU	Intel(R) Xeon(R) Gold 6326 CPU @ 2.90GHz
GPU	2× NVIDIA A10
Operating system	Ubuntu 20.04.6 LTS
CUDA version	CUDA 10.1

Table 1: Experimental configuration for training and validation processes.

Hyperparameter	Value
Optimizer	SGD with Nesterov momentum
Learning rate	0.01
Momentum	0.9
Loss weights (Bbox, Cls, DFL)	(7.5, 0.5, 1.5)
Maximum epochs	1000 (+ early stopping)
Patience	50

Table 2: Hyperparameter settings for every model training and validation.

Augmentation	Precision	Recall	mAP 50	mAP 50:95
Rotation(90º,180º,270º)	0.744	0.748	0.8	0.571
Flipping(horizontally, vertically)	0.795	0.731	0.802	0.554
Random cropping	0.834	0.712	0.782	0.556
Colour enhancement	0.792	0.744	0.806	0.57
Histogram equalization	0.779	0.721	0.785	0.555
Random occlusion	0.825	0.679	0.788	0.544

Table 3: Training metrics for COTONET on augmented CONDIS++ datasets, given artificially augmented techniques.

Model	Model Size (M)	GFLOPs	Precision	Recall	mAP 50	mAP 50:95
YOLOv8n	3.1	8.9	0.769	0.707	0.762	0.553
YOLOv8s	11.1	28.8	0.800	0.697	0.771	0.566
YOLOv8m	25.9	79.3	0.819	0.725	0.807	0.604
YOLOv9t	2.1	8.5	0.771	0.755	0.791	0.566
YOLOv9s	7.3	27.6	0.710	0.761	0.784	0.585
YOLOv9m	20.2	77.9	0.804	0.720	0.795	0.603

Table 4: Training results on augmented CONDIS++ for the proposed algorithm and state-of-the-art YOLO baseline models.

Attention	Added parameters	GFLOPs	Precision	Recall	mAP 50	mAP 50:95
No attention	0	27.8	0.809	0.763	0.817	0.581
ECA	3	27.8	0.78	0.742	0.814	0.587
SE	19.854	27.8	0.825	0.733	0.811	0.606

Table 5: Lightweight attention operators comparative analysis after backbone’s convolutional blocks for COTONET.

Attention	Added parameters	GFLOPs	Precision	Recall	mAP 50	mAP 50:95
No attention	0	27.5	0.751	0.747	0.795	0.574
SimAM	0	27.5	0.82	0.728	0.803	0.58
PHAM	126196	27.8	0.787	0.729	0.815	0.572
SimAM + PHAM	126196	27.8	0.825	0.733	0.811	0.606

Table 6: Ablation study of attention mechanisms in COTONET.

Attention	Added parameters	GFLOPs	Precision	Recall	mAP 50	mAP 50:95
CBAM	657740	28.1	0.815	0.735	0.808	0.587
EMA	6624	28.9	0.789	0.781	0.825	0.603
MAA	2153440	42.3	0.714	0.772	0.807	0.58
MCA	98	27.6	0.739	0.757	0.798	0.57
TripletAttn	1200	28.4	0.83	0.721	0.799	0.571
SimAM + PHAM	126196	27.8	0.825	0.733	0.811	0.606

Table 7: Comparative analysis of different attention mechanisms in COTONET.

Panel of Figure 11 — Figure 11: Detection comparison between standard YOLO models and COTONET. Column-wise, left to right: YOLOv8m, YOLOv9m, YOLOv10m, YOLO11m, YOLOv12m, and COTONET predictions over different CONDIS++ classes.

Bounding box loss function	Precision	Recall	mAP 50	mAP 50:95
CIoU	0.828	0.686	0.809	0.561
DIoU	0.809	0.700	0.799	0.556
GIoU	0.818	0.734	0.817	0.580
MDPIoU	0.776	0.737	0.791	0.567
FocalerIoU	0.783	0.733	0.809	0.586
SIoU	0.828	0.733	0.811	0.606

Table 8: Comparative analysis of different bounding box loss functions in COTONET.

Object Detection

Object detection locates and classifies objects in images or video, balancing localization accuracy, robustness, and inference speed.

See all 624 papers

Precision Cotton Disease Management: Leveraging UAV Imagery and Custom YOLO Deep Learning Framework for Enhanced Detection
Oct 16, 20261. Anwar Iqbal*1 aiqbal. rime23smme@student. nust. edu. pk 2. Kunwar Faraz Ahmad Khan1 kunwar. faraz@smme. nust. edu. pk 3. Shahbaz Khan1 shahbaz. khan@smme. nust. edu. pk 4. Khawaja Fahad Iqbal1 fahad. iqbal@smme. nust. edu. pk 5. Sara Ali1, 2 sara. baber@smme. nust. edu. pk 6. Saqib Nazir1 saqib. nazir@smme. nust. edu. pk 7. Muhammad Azhar Javed3 muhammadazhar6231@gmail. com 1 Department of Robotics, Artificial Intelligence +11Efficient Object Detection Plant Disease Classification
NUST
Selective Cotton Boll Localization for Robotic Harvesting: Evaluation of Deep Learning Vision Models Under Field Conditions
Sep 17, 2026Thevathayarajh Thayananthan, Xin Zhang, Isuru Laddusinghe Badu +6Robotic Harvesting Instance Segmentation
School of Environmental, Civil, Agricultural and Mechanical Engineering, University of Georgia, Athens, 30602, GA, USA · Department of Entomology, University of Georgia, Tifton, 31793, GA, USA · Department of Crop and Soil Sciences, University of Georgia, Athens, 30602, GA, USA +1
Cotton-SF YOLO: Learning Structural and Frequency Cues for Early Cotton Square Detection in Complex Field Environments
Jul 16, 2026Chengjia Zhang, Yu Li, Feiri Ali +5Agricultural Image Analysis Frequency-Domain Feature Learning
School of Aeronautics and Astronautics, Xichang University, No. 1 Xuefu Road, Anning Town, Xichang, 615013, Sichuan Province, China · School of Information Technology, Xichang University, No. 1 Xuefu Road, Anning Town, Xichang, 615013, Sichuan Province, China · College of Information and Electrical Engineering, China Agricultural University, No. 17 Qinghua East Road, Haidian District, Beijing, 100083, China +1

Oct 16, 2026J/K move · Enter open · S save

Precision Cotton Disease Management: Leveraging UAV Imagery and Custom YOLO Deep Learning Framework for Enhanced Detection

1. Anwar Iqbal*1 aiqbal. rime23smme@student. nust. edu. pk 2. Kunwar Faraz Ahmad Khan1 kunwar. faraz@smme. nust. edu. pk 3. Shahbaz Khan1 shahbaz. khan@smme. nust. edu. pk 4. Khawaja Fahad Iqbal1 fahad. iqbal@smme. nust. edu. pk 5. Sara Ali1, 2 sara. baber@smme. nust. edu. pk 6. Saqib Nazir1 saqib. nazir@smme. nust. edu. pk 7. Muhammad Azhar Javed3 muhammadazhar6231@gmail. com 1 Department of Robotics, Artificial Intelligence, SMME+10

NUST

Efficient Object Detection Plant Disease Classification Precision Agriculture Aerial Object Detection

Open paper AskPDF arXiv

Cotton sustains the textile economies of over 80 countries with an annual yield of 113 million bales, yet 30-60% of yield losses are still attributed to viral, bacterial, and fungal diseases that remain undetected until manual field scouting. To address this critical issue, the integration of Unmanned Aerial Vehicle (UAV) imagery with deep learning offers a scalable early-warning system; however, existing models suffer from low early-stage accuracy, poor transferability, and high computational cost. In this paper, we propose Cotton-Vision, a fully convolutional single-stage detector that synergizes three lightweight architectural innovations: Ghost Bottleneck modules for a 20% parameter reduction, Efficient Channel Attention (ECA) for adaptive feature recalibration, and Cross-Stage Partial (CSP) fusion for enriched gradient flow. The network is trained on a novel dataset collected across seven cotton fields in the Punjab province of Pakistan, comprising over 2,700 high-resolution drone images. Extensive data augmentation (rotation, shear, exposure, and saturation) expands the dataset to 11,704 training samples. On a held-out test set, Cotton-Vision achieves a precision of 97.7%, a recall of 77.7%, an mAP@50 of 85.6%, and an mAP@50-95 of 76.6%. These results represent a 10-20% improvement in precision, recall, and mAP over state-of-the-art baselines, including YOLOv5, YOLOv8, YOLOv11, YOLOv12, RT-DETR, and RF-DETR. Through a comprehensive ablation study, we demonstrate that each proposed architectural enhancement contributes to the model's superior performance, yielding a 12-16% mAP improvement over the baseline architecture. Our findings confirm that this synergy of lightweight and attention-based modules provides a robust, computationally viable solution for early cotton disease detection.

Abstract

Object Detection

Explore similar work

Precision Cotton Disease Management: Leveraging UAV Imagery and Custom YOLO Deep Learning Framework for Enhanced Detection

Selective Cotton Boll Localization for Robotic Harvesting: Evaluation of Deep Learning Vision Models Under Field Conditions

Cotton-SF YOLO: Learning Structural and Frequency Cues for Early Cotton Square Detection in Complex Field Environments