Introduction: Effective pest management is important during the natural growth phases of cotton in the wild. As cotton fields are infested with “tiny pests” (smaller than 32×32 pixels) and “very tiny pests” (smaller than 16×16 pixels) during growth, making it difficult for common object detection models to accurately detect and fail to make sound agricultural decisions.Methods: In this study, we proposed a framework for detecting “tiny pests” and “very tiny pests” in wild cotton fields, named SRNet-YOLO. SRNet-YOLO includes a YOLOv8 feature extraction module, a feature map super-resolution reconstruction module (FM-SR), and a fusion mechanism based on BiFormer attention (BiFormerAF). Specially, the FM-SR module is designed for the feature map level to recover the important feature in detail, in other words, this module reconstructs the P5 layer feature map into the size of the P3 layer. And then we designed the BiFormerAF module to fuse this reconstruct layer with the P3 layer, which greatly improves the detection performance. The purpose of the BiFormerAF module is to solve the problem of possible loss of feature after reconstruction. Additionally, to validate the performance of our method for “tiny pests” and “very tiny pests” detection in cotton fields, we have developed a large dataset, named Cotton-Yellow-Sticky-2023, which collected pests by yellow sticky traps.Results: Through comprehensive experimental verification, we demonstrate that our proposed framework achieves exceptional performance. Our method achieved 78.2% mAP on the “tiny pests” test result, it surpasses the performance of leading detection models such as YOLOv3, YOLOv5, YOLOv7 and YOLOv8 by 6.9%, 7.2%, 5.7% and 4.1%, respectively. Meanwhile, our results on “very tiny pests” reached 57% mAP, which are 32.2% higher than YOLOv8. To verify the generalizability of the model, our experiments on Yellow Sticky Traps (low-resolution) dataset still maintained the highest 92.8% mAP.Discussion: The above experimental results indicate that our model not only provides help in solving the problem of tiny pests in cotton fields, but also has good generalizability and can be used for the detection of tiny pests in other crops.

Deep learning-based salient object detection (SOD) methods have made great progress in recent years. However, most deep learning-based methods suffer from coarse object boundaries and expensive computations, especially in detecting objects with complex shapes. This paper presents an accurate and efficient SOD method that is based on a novel double-branch network that includes a body branch and an edge branch. To obtain an accurate edge, an edge profile enhancement module (EPEM) is embedded in the edge branch. In addition, a fusion feedback module (FFM) is embedded to integrate features from the two branches. To address the problem of expensive computations, channel attention module (CAM) is included to restrain redundant feature channels. Thus, the speed of the inference step can be improved with little reduction in the boundary accuracy. Experimental results on 9 datasets demonstrate that the proposed method performs favorably against 8 state-of-the-art methods in terms of both accuracy and efficiency. Additionally, our method achieves excellent detection results for objects with complex shapes.

Adverse weather conditions can significantly hinder the performance of deep learning-based object detection models. Traditional approaches often rely on image restoration techniques to enhance the quality of degraded images prior to detection. However, these methods frequently struggle to balance image enhancement and detection tasks effectively, often overlooking latent information that could be beneficial for detection. To address these challenges, we propose a novel framework: Knowledge Distillation based on Diffusion Models (KDDM). This framework incorporates a Dehaze Network (DN), which employs large kernel convolution to remove weather-specific artifacts, thereby revealing more latent information. The DN, together with a text detector, forms an end-to-end scene text detection network, acting as the student network. Additionally, the nuanced internal representations of text-to-image diffusion models adeptly capture and integrate higher-order visual semantic concepts. Given the rich textual and visual content inherent in scene text, there is a fundamental connection to text-to-image diffusion models. As such, we utilize diffusion models as a teacher network to distill high-level visual semantic knowledge into the student network. Notably, we introduce an innovative distillation technique using a “Threshold_Mask”, which ensures that the student network focuses on text regions while minimizing interference from irrelevant background elements. Comprehensive experimental evaluations demonstrate that our KDDM framework significantly outperforms baseline models under foggy weather conditions, marking a substantial advancement in the field.

Pest detection in the field is crucial for realizing smart agriculture. Deep learning-based target detection algorithms have become an important pest identification method due to their high detection accuracy, but the existing methods still suffer from misdetection and omission when detecting small-targeted pests and small-targeted pests in more complex backgrounds. For this reason, this study improves on YOLO11 and proposes a new model called MSDS-YOLO for enhanced detection of small-target pests. First, a new dynamic multi-scale feature extraction module (C3k2_DMSFE) is introduced, which can be adaptively adjusted according to different input features and thus effectively capture multi-scale and diverse feature information. Next, a novel Dimensional Selective Feature Pyramid Network (DSFPN) is proposed, which employs adaptive feature selection and multi-dimensional fusion mechanisms to enhance small-target saliency. Finally, the ability to fit small targets was enhanced by adding 160 × 160 detection heads removing 20 × 20 detection heads and using Normalized Gaussian Wasserstein Distance (NWD) combined with CIoU as a position loss function to measure the prediction error. In addition, a real small-target pest dataset, Cottonpest2, is constructed for validating the proposed model. The experimental results showed that a mAP50 of 86.7% was achieved on the self-constructed dataset Cottonpest2, which was improved by 3.0% compared to the baseline. At the same time, MSDS-YOLO has achieved better detection accuracy than other YOLO models on public datasets. Model evaluation on these three datasets shows that the MSDS-YOLO model has excellent robustness and model generalization ability.