Search

Abstract The synthesis of facial sketch-photo has important applications in practical life, such as crime investigation. Many convolutional neural networks (CNNs) based methods have been proposed to address this issue. However, due to the substantial modal differences between sketch and photo, the CNN’s insensitivity to global information, and insufficient utilization of hierarchical features, synthesized photos struggle to balance both identity preservation and image quality. Recently, State Space Sequence Models (SSMs) have achieved exciting results in computer vision (CV) tasks. Inspired by SSMs, we design a hybrid CNN–SSM model called FaceMamba for the Face Sketch-Photo Synthesis (FSPS) task. It includes an original Face Vision Mamba Attention for modeling in latent space using SSM. Additionally, it incorporates a general auxiliary method called Attention Feature Injection that combines encoding features, decoding features, and external auxiliary features using attention mechanisms. FaceMamba combines Mamba’s modeling ability for long-range dependencies with CNN’s powerful local feature extraction ability, and utilizes hierarchical features at the appropriate position. Adequate experimental and evaluation results reveal that FaceMamba has strong competitiveness in FSPS task, achieving the best balance between identity preservation and image quality.

Frequent pest infestations in rice can substantially decrease rice yield, severely hindering agricultural production. Computer vision can assist farmers in timely locating and removing partial or even entire plants infested with pests, achieving pest control. However, the size and population density variance of paddy pests commonly cause interclass and intraclass differences in pest datasets, as well as an imbalance in hard samples. In addition, existing deep learning models struggle to balance compactness with high detection accuracy. This paper addresses these challenges by constructing the GPest14 dataset using FastGAN image generation technology, web crawlers, and manual selection. This is a large-scale rice pest dataset containing 14 categories suitable for training fine-grained visual detection tasks of rice pests. This work proposes the Fully Connected Bottleneck Transformer (FCBT) module, implants it into the tail of the YOLOv8n architecture backbone, and constructs the FCBTYOLO object detection model. Results indicate that the Gpest14 dataset increases by at least 12.2% in mean Average Precision (mAP) across major detection models. Simultaneously, our proposed FCBTYOLO model can achieve an average precision (mAP@50) of 93.6% with the network training weights constrained to a mere 6.7MB, and it only requires 16.8 milliseconds to detect a single pest image. FCBTYOLO has been validated on the large-scale public Pest24 dataset, where it can achieve a 1.1% improvement in mAP over the YOLOv8n model while maintaining the training weight at 8.3MB. Compared with existing pest detection models, FCBTYOLO exhibits lighter, superior performance and is more suitable for field-embedded deployment in digital agriculture.