Search

Abstract Highly porous metal-organic framework (MOF) nanosheets have shown promising potential for efficient water sorption kinetics in atmospheric water harvesting (AWH) systems. However, the water uptake of single-component MOF absorbents remains limited due to their low water retention. To overcome this limitation, we present a strategy for fabricating vertically aligned MOF nanosheets on hydrogel membrane substrates (MOF-CT/PVA) to achieve ultrafast AWH with high water uptake. By employing directional growth of MOF nanosheets, we successfully create superhydrophilic MOF coating layer and pore channels for efficient water transportation to the crosslinked flexible hydrogel membrane. The designed composite water harvester exhibits ultrafast sorption kinetics, achieving 91.4% saturation within 15 min. Moreover, MOF-CT/PVA exhibits superior solar-driven water capture-release capacity even after 10 cycles of reuse. This construction approach significantly enhances the water vapor adsorption, offering a potential solution for the design of composite MOF-membrane harvesters to mitigate the freshwater crisis.

Abstract The rapid development of optical frequency combs from their table-top origins towards chip-scale platforms has opened up exciting possibilities for comb functionalities outside laboratories. Enhanced nonlinear processes in microresonators have emerged as a mainstream comb-generating mechanism with compelling advantages in size, weight, and power consumption. The established understanding of gain and loss in nonlinear microresonators, along with recently developed ultralow-loss nonlinear photonic circuitry, has boosted the optical energy conversion efficiency of microresonator frequency comb (microcomb) devices from below a few percent to above 50%. This review summarizes the latest advances in novel photonic devices and pumping strategies that contribute to these milestones of microcomb efficiency. The resulting benefits for high-performance integration of comb applications are also discussed before summarizing the remaining challenges.

Graph convolution subspace clustering has been widely used in the field of hyperspectral image (HSI) unsupervised classification due to its ability to aggregate neighborhood information. However, existing methods focus on using graph convolution techniques to design feature extraction functions, ignoring the mutual optimization of the graph convolution operator and the self-expression coefficient matrix, leading to suboptimal clustering results. In addition, these methods directly construct graphs on raw data, which may be easily affected by noises and then degrade the clustering performance, as the constructed topology is not credible for the training procedure. To address these issues, we propose a novel method called spatial-spectral adaptive graph convolutional subspace clustering (S2AGCSC). We employ the reconstruction coefficient matrix to devise a graph convolutional operator with adjacency matrix, which collaboratively computes both the feature representations and coefficient matrix, and the graph-convolutional operator is updated iteratively and adaptively during training. In addition, we harness a combination of spectral and spatial features to introduce additional view information to help learn more robust features and generate more refined superpixels. Experimental validation on three HSI datasets confirms the efficacy of S2AGCSC.

Abstract Dissipative Kerr solitons from optical microresonators, commonly referred to as soliton microcombs, have been developed for a broad range of applications, including precision measurement, optical frequency synthesis, and ultra-stable microwave and millimeter wave generation, all on a chip. An important goal for microcombs is self-referencing, which requires octave-spanning bandwidths to detect and stabilize the comb carrier envelope offset frequency. Further, detection and locking of the comb spacings are often achieved using frequency division by electro-optic modulation. The thin-film lithium niobate photonic platform, with its low loss, strong second- and third-order nonlinearities, as well as large Pockels effect, is ideally suited for these tasks. However, octave-spanning soliton microcombs are challenging to demonstrate on this platform, largely complicated by strong Raman effects hindering reliable fabrication of soliton devices. Here, we demonstrate entirely connected and octave-spanning soliton microcombs on thin-film lithium niobate. With appropriate control over microresonator free spectral range and dissipation spectrum, we show that soliton-inhibiting Raman effects are suppressed, and soliton devices are fabricated with near-unity yield. Our work offers an unambiguous method for soliton generation on strongly Raman-active materials. Further, it anticipates monolithically integrated, self-referenced frequency standards in conjunction with established technologies, such as periodically poled waveguides and electro-optic modulators, on thin-film lithium niobate.

Abstract Resonator-based optical frequency comb generation is an enabling technology for a myriad of applications ranging from communications to precision spectroscopy. These frequency combs can be generated in nonlinear resonators driven using either continuous-wave (CW) light, which requires alignment of the pump frequency with the cavity resonance, or pulsed light, which also mandates that the pulse repetition rate and cavity free spectral range (FSR) are carefully matched. Advancements in nanophotonics have ignited interest in chip-scale optical frequency combs. However, realizing pulse-driven on-chip Kerr combs remains challenging, as microresonator cavities have limited tuning range in their FSR and resonance frequency. Here, we take steps to overcome this limitation and demonstrate broadband frequency comb generation using a χ(3) resonator synchronously pumped by a tunable femtosecond pulse generator with on-chip amplitude and phase modulators. Notably, employing pulsed pumping overcomes limitations in Kerr comb generation typically seen in crystalline resonators from stimulated Raman scattering.

Background Endovascular treatment (EVT) has been demonstrated to be effective for patients with tandem occlusion. The efficacy and safety of intravenous thrombolysis before EVT in patients with tandem occlusion remain debatable. Methods and Results We conducted a systematic review and meta‐analysis with PubMed, EMBASE, and the Cochrane Library from inception to September 2023. The primary outcome was functional independence, defined as a modified Rankin Scale score of 0 to 2 at 90 days. The secondary outcomes included the successful recanalization rate, symptomatic intracerebral hemorrhage, and mortality at 90 days. In total, 9 studies with 1838 enrolled participants were identified. Our results showed that, compared with treatment with EVT alone, intravenous thrombolysis before EVT was associated with a greater proportion of functional independence at 90 days (odds ratio [OR], 1.39 [95% CI, 1.14–1.69]; P=0.001), a greater rate of successful recanalization (OR, 1.45 [95% CI, 1.11–1.89]; P=0.007) and decreased mortality (OR, 0.68 [95% CI, 0.50–0.93]; P=0.02). Furthermore, there was no significant difference in symptomatic intracerebral hemorrhage between the intravenous thrombolysis plus EVT group and the EVT alone group (OR, 1.16 [95% CI, 0.79–1.70]; P=0.45). Conclusions In patients with acute ischemic stroke and tandem occlusion, intravenous thrombolysis before EVT was associated with a greater rate of favorable functional outcomes and successful recanalization and a lower mortality rate without an increased risk of symptomatic intracerebral hemorrhage compared with patients receiving EVT alone.

PD-1 blockade therapy has made great breakthroughs in treatment of multiple solid tumors. However, patients with microsatellite-stable (MSS) colorectal cancer (CRC) respond poorly to anti-PD-1 immunotherapy. Although CRC patients with microstatellite instability (MSI) or microsatellite instability-high (MSI-H) can benefit from PD-1 blockade therapy, there are still some problems such as tumor recurrence. Tumor-associated macrophages (TAMs), most abundant immune components in tumor microenvironment (TME), largely limit the therapeutic efficacy of anti-PD-1 against CRC. The CSF1/CSF1R pathway plays a key role in regulating macrophage polarization, and blocking CSF1R signaling transduction may be a potential strategy to effectively reprogram macrophages and remodel TME. Here, we found that increasing expression of CSF1R in macrophages predicted poor prognosis in CRC cohort. Furthermore, we discovered a novel potent CSF1R inhibitor, PXB17, which significantly reprogramed M2 macrophages to M1 phenotype. Mechanically, PXB17 significantly blocked activation of PI3K/AKT/mTORC1 signaling, resulting in inhibition of cholesterol biosynthesis. Results from 3D co-culture system suggested that PXB17-repolarized macrophages could induce infiltration of CD8+ T lymphocytes in tumors and improve the immunosuppressive microenvironment. In vivo, PXB17 significantly halted CRC growth, with a stronger effect than PLX3397. In particular, PXB17 potently enhanced therapeutic activity of PD-1 mAb in CT-26 (MSS) model and prevented tumor recurrence in MC-38 (MSI-H) model by promoting formation of long-term memory immunity. Our study opens a new avenue for CSF1R in tumor innate and adaptive anti-tumor immunomodulatory activity and suggests that PXB17 is a promising immunotherapy molecule for enhancing the efficacy of PD-1 mAb or reducing tumor recurrence of CRC.

Accurate segmentation of the stem of pumpkin seedlings has a great influence on the modernization of pumpkin cultivation, and can provide detailed data support for the growth of pumpkin plants. We collected and constructed a pumpkin seedling point cloud dataset for the first time. Potting soil and wall background in point cloud data often interfere with the accuracy of partial cutting of pumpkin seedling stems. The stem shape of pumpkin seedlings varies due to other environmental factors during the growing stage. The stem of the pumpkin seedling is closely connected with the potting soil and leaves, and the boundary of the stem is easily blurred. These problems bring challenges to the accurate segmentation of pumpkin seedling point cloud stems. In this paper, an accurate segmentation algorithm for pumpkin seedling point cloud stems based on CPHNet is proposed. First, a channel residual attention multilayer perceptron (CRA-MLP) module is proposed, which suppresses background interference such as soil. Second, a position-enhanced self-attention (PESA) mechanism is proposed, enabling the model to adapt to diverse morphologies of pumpkin seedling point cloud data stems. Finally, a hybrid loss function of cross entropy loss and dice loss (HCE-Dice Loss) is proposed to address the issue of fuzzy stem boundaries. The experimental results show that CPHNet achieves a 90.4% average cross-to-merge ratio (mIoU), 93.1% average accuracy (mP), 95.6% average recall rate (mR), 94.4% F1 score (mF1) and 0.03 plants/second (speed) on the self-built dataset. Compared with other popular segmentation models, this model is more accurate and stable for cutting the stem part of the pumpkin seedling point cloud.

BackgroundThe lack of randomized evidence makes it difficult to establish reliable treatment recommendations for endovascular treatment (EVT) in elderly patients. This meta-analysis aims to evaluate the therapeutic effects of endovascular treatment for acute ischemic stroke in the elderly compared with younger patients.MethodsComprehensive literature retrieval was conducted to identify studies that directly compared the outcomes of EVT in elderly patients and those aged

Abstract Background There is a paucity of data regarding the prevalence of preoperative deep vein thrombosis (DVT) in patients with long bone (including femur, tibia and fibula) fractures of the lower limbs. We performed a meta-analysis to address the issue. Methods Electronic databases, including PubMed, EMBASE, the Web of Science, the Cochrane Library, the VIP database, CNKI, and the Wanfang database, were systematic searched for original articles that reported the prevalence of preoperative DVT in long bone fractures of the lower limbs from January 2016 to September 2021. The prevalence of preoperative DVT was pooled using random-effects models, and subgroups were established according to study type, detection method, sample size and fracture site. Results Twenty-three articles reporting on 18,119 patients were eligible. The overall pooled preoperative DVT prevalence was 24.1% (95% CI 19.3–28.8%). In different subgroups, the preoperative DVT prevalences were 18.2–27.3%, 15.2–28.6%, 23.1–24.9%, 18.2–26.0% and 23.2–23.4% for different study designs, sample sizes, age groups, detection methods and fracture sites, respectively. Conclusions Despite the heterogeneity among studies, this systematic review suggests that the prevalence of preoperative DVT, which may seriously affect the prognosis of patients, is high. Therefore, greater efforts should be devoted to the improvement of screening and prevention strategies for preoperative DVT in lower-extremity long bone fractures. Level of Evidence: Level III. Trial Registration The study was registered in the International Prospective Register of Systematic Reviews (PROSPERO) database with the registration number CRD42022324706.

Electro-optic modulators can be useful for imaging, sensing and information processing applications. Here the authors demonstrate an ultra-low drive voltage visible to near infrared range electro-optic modulator in the form of amplitude and phase modulation using thin-film lithium niobate.

Background & Aims: The machinery that prevents colorectal cancer liver metastasis (CRLM) in the context of liver regeneration (LR) remains elusive. Ceramide (CER) is a potent anti-cancer lipid involved in intercellular interaction. Here, we investigated the role of CER metabolism in mediating the interaction between hepatocytes and metastatic colorectal cancer (CRC) cells to regulate CRLM in the context of LR. Methods: Mice were intrasplenically injected with CRC cells. LR was induced by 2/3 partial hepatectomy (PH) to mimic the CRLM in the context of LR. The alteration of corresponding CER-metabolizing genes was examined. The biological roles of CER metabolism in vitro and in vivo were examined by performing a series of functional experiments. Results: Induction of LR augmented apoptosis but promoted matrix metalloproteinase 2 (MMP2) expression and epithelial-mesenchymal transition (EMT) to increase the invasiveness of metastatic CRC cells, resulting in aggressive CRLM. Up-regulation of sphingomyelin phosphodiesterase 3 (SMPD3) was determined in the regenerating hepatocytes after LR induction and persisted in the CRLM-adjacent hepatocytes after CRLM formation. Hepatic Smpd3 knockdown was found to further promote CRLM in the context of LR by abolishing mitochondrial apoptosis and augmenting the invasiveness in metastatic CRC cells by up-regulating MMP2 and EMT through promoting the nuclear translocation of β-catenin. Mechanistically, we found that hepatic SMPD3 controlled the generation of exosomal CER in the regenerating hepatocytes and the CRLM-adjacent hepatocytes. The SMPD3-produced exosomal CER critically conducted the intercellular transfer of CER from the hepatocytes to metastatic CRC cells and impeded CRLM by inducing mitochondrial apoptosis and restricting the invasiveness in metastatic CRC cells. The administration of nanoliposomal CER was found to suppress CRLM in the context of LR substantially. Conclusions: SMPD3-produced exosomal CER constitutes a critical anti-CRLM mechanism in LR to impede CRLM, offering the promise of using CER as a therapeutic agent to prevent the recurrence of CRLM after PH.

An integrated thin-film lithium niobate phase modulator enables on-chip frequency and bandwidth control of single photons

We show frequency domain mirrors that provide reflections of optical mode propagation in the frequency domain. We theoretically investigated the mirror properties and experimentally demonstrate it using polarization and coupled-resonator-based coupling on thin film Lithium Niobate.

Summary: Patient-derived tumor organoids (PDTOs) have the potential to be used to predict the patient response to chemotherapy. However, the cutoff value of the half-maximal inhibition concentration (IC50) for PDTO drug sensitivity has not been validated with clinical cohort data. We established PDTOs and performed a drug test in 277 samples from 242 CRC patients who received FOLFOX or XELOX chemotherapy. After follow-up and comparison of the PDTO drug test and final clinical outcome results, the optimal IC50 cutoff value for PDTO drug sensitivity was 43.26 μmol/L. This PDTO drug test-defined cutoff value could predict patient response with 75.36% sensitivity, 74.68% specificity, and 75% accuracy. Moreover, this value distinguished groups of patients with significant differences in survival benefit. Our study is the first to define the IC50 cutoff value for the PDTO drug test to effectively distinguish CRC patients with chemosensitivity or nonsensitivity and predict survival benefits.

Summary: The tumor, nodes and metastasis (TNM) classification system provides useful but incomplete prognostic information and lacks the assessment of the tumor microenvironment (TME). Collagen, the main component of the TME extracellular matrix, plays a nonnegligible role in tumor invasion and metastasis. In this cohort study, we aimed to develop and validate a TME collagen signature (CSTME) for prognostic prediction of stage II/III colorectal cancer (CRC) and to compare the prognostic values of “TNM stage + CSTME” with that of TNM stage alone. Results indicated that the CSTME was an independent prognostic risk factor for stage II/III CRC (hazard ratio: 2.939, 95% CI: 2.180–3.962, p

Graph convolutional networks (GCNs) are a promising approach for addressing the necessity for long-range information in hyperspectral image (HSI) classification. Researchers have attempted to develop classification methods that combine strong generalizations with effective classification. However, the current HSI classification methods based on GCN present two main challenges. First, they overlook the multi-view features inherent in HSIs, whereas multi-view information interacts with each other to facilitate classification tasks. Second, many algorithms perform a rudimentary fusion of extracted features, which can result in information redundancy and conflicts. To address these challenges and exploit the strengths of multiple features, this paper introduces an adaptive multi-feature fusion GCN (AMF-GCN) for HSI classification. Initially, the AMF-GCN algorithm extracts spectral and textural features from the HSIs and combines them to create fusion features. Subsequently, these three features are employed to construct separate images, which are then processed individually using multi-branch GCNs. The AMG-GCN aggregates node information and utilizes an attention-based feature fusion method to selectively incorporate valuable features. We evaluated the model on three widely used HSI datasets, i.e., Pavia University, Salinas, and Houston-2013, and achieved accuracies of 97.45%, 98.03%, and 93.02%, respectively. Extensive experimental results show that the classification performance of the AMF-GCN on benchmark HSI datasets is comparable to those of state-of-the-art methods.

Affected by various environmental factors, citrus will frequently suffer from diseases during the growth process, which has brought huge obstacles to the development of agriculture. This paper proposes a new method for identifying and classifying citrus diseases. Firstly, this paper designs an image enhancement method based on the MSRCR algorithm and homomorphic filtering algorithm optimized by Laplacian (HFLF-MS) to highlight the disease characteristics of citrus. Secondly, we designed a new neural network DS-MENet based on the DenseNet-121 backbone structure. In DS-MENet, the regular convolution in Dense Block is replaced with depthwise separable convolution, which reduces the network parameters. The ReMish activation function is used to alleviate the neuron death problem caused by the ReLU function and improve the robustness of the model. To further enhance the attention to citrus disease information and the ability to extract feature information, a multi-channel fusion backbone enhancement method (MCF) was designed in this work to process Dense Block. We use the 10-fold cross-validation method to conduct experiments. The average classification accuracy of DS-MENet on the dataset after adding noise can reach 95.02%. This shows that the method has good performance and has certain feasibility for the classification of citrus diseases in real life.

IntroductionRetroform cervical dystonia (RCD), which includes retrocaput and retrocollis, is a rare form of cervical dystonia. Few reports have been published on RCD. The present study aimed to characterize the target muscles involved in RCD and the efficacy of botulinum toxin type A (BTX-A) injection.MethodsPatients with consecutive cervical dystonia with RCD as the most problematic feature were retrospectively analyzed over a 10-year period. Target muscles were screened and confirmed based on clinical evaluation, single-photon emission computed tomography, and electromyography. In addition, efficacy and adverse events following BTX-A injection in patients with RCD were evaluated.ResultsA total of 34 patients with RCD were included, 18 of whom presented with retrocaput and 16 with retrocollis. The most frequently injected muscles in RCD were splenius capitis (SPCa, 97.1%) and semispinalis capitis (SSCa, 97.1%), followed by levator scapulae (LS, 50.0%), rectus capitis posterior major (RCPM, 47.1%), trapezius (TPZ, 41.2%), and sternocleidomastoid muscle (SCM, 41.2%). Besides cervical muscles, the erector spinae was also injected in 17.6% of patients. Most muscles were predominantly bilaterally injected. The injection schemes of retrocaput and retrocollis were similar, possibly because in patients with retrocollis, retrocaput was often combined. BTX-A injection achieved a satisfactory therapeutic effect in RCD, with an average symptom relief rate of 69.0 ± 16.7%. Mild dysphagia (17.6%) and posterior cervical muscle weakness (17.6%) were the most common adverse events.ConclusionSPCa, SSCa, LS, RCPM, LS, and SCM were commonly and often bilaterally injected in RCD. Patients with RCD could achieve satisfactory symptom relief after BTX-A injection.

Grape disease is a significant contributory factor to the decline in grape yield, typically affecting the leaves first. Efficient identification of grape leaf diseases remains a critical unmet need. To mitigate background interference in grape leaf feature extraction and improve the ability to extract small disease spots, by combining the characteristic features of grape leaf diseases, we developed a novel method for disease recognition and classification in this study. First, Gaussian filters Sobel smooth de-noising Laplace operator (GSSL) was employed to reduce image noise and enhance the texture of grape leaves. A novel network designated coordinated attention shuffle mechanism-asymmetric multi-scale fusion module net (CASM-AMFMNet) was subsequently applied for grape leaf disease identification. CoAtNet was employed as the network backbone to improve model learning and generalization capabilities, which alleviated the problem of gradient explosion to a certain extent. The CASM-AMFMNet was further utilized to capture and target grape leaf disease areas, therefore reducing background interference. Finally, Asymmetric multi-scale fusion module (AMFM) was employed to extract multi-scale features from small disease spots on grape leaves for accurate identification of small target diseases. The experimental results based on our self-made grape leaf image dataset showed that, compared to existing methods, CASM-AMFMNet achieved an accuracy of 95.95%, F1 score of 95.78%, and mAP of 90.27%. Overall, the model and methods proposed in this report could successfully identify different diseases of grape leaves and provide a feasible scheme for deep learning to correctly recognize grape diseases during agricultural production that may be used as a reference for other crops diseases.

In deep learning-based maize leaf disease detection, a maize disease identification method called Network based on wavelet threshold-guided bilateral filtering, multi-channel ResNet, and attenuation factor (WG-MARNet) is proposed. This method can solve the problems of noise, background interference, and low detection accuracy of maize leaf disease images. To begin, a processing layer called Wavelet threshold guided bilateral filtering (WT-GBF) based on the WG-MARNet model is employed to reduce image noise and perform high and low-frequency decomposition of the input image using WT-GBF. This increases the input image's resistance to environmental interference and feature extraction capability. Secondly, for the multiscale feature fusion technique, an average down-sampling and tiling method is employed to increase feature representation and limit the risk of overfitting. Then, on high and low-frequency multi-channel, an attenuation factor is introduced to optimize the performance instability during training of the deep network. Finally, when the convergence and accuracy are compared, PRelu and Adabound are used instead of the Relu activation function and the Adam optimizer. The experimental results revealed that our method's average recognition accuracy was 97.96%, and the detection time for a single image was 0.278 seconds. The average detection accuracy has been increased. The method lays the groundwork for the precise control of maize diseases in the field.

The appearance quality of apples directly affects their price. To realize apple grading automatically, it is necessary to find an effective method for detecting apple surface defects. Aiming at the problem of a low recognition rate in apple surface defect detection under small sample conditions, we designed an apple surface defect detection network (ASDINet) suitable for small sample learning. The self-developed apple sorting system collected RGB images of 50 apple samples for model verification, including non-defective and defective apples (rot, disease, lacerations, and mechanical damage). First, a segmentation network (AU-Net) with a stronger ability to capture small details was designed, and a Dep-conv module that could expand the feature capacity of the receptive field was inserted in its down-sampling path. Among them, the number of convolutional layers in the single-layer convolutional module was positively correlated with the network depth. Next, to achieve real-time segmentation, we replaced the flooding of feature maps with mask output in the 13th layer of the network. Finally, we designed a global decision module (GDM) with global properties, which inserted the global spatial domain attention mechanism (GSAM) and performed fast prediction on abnormal images through the input of masks. In the comparison experiment with state-of-the-art models, our network achieved an AP of 98.8%, and a 97.75% F1-score, which were higher than those of most of the state-of-the-art networks; the detection speed reached 39ms per frame, achieving accuracy-easy deployment and substantial trade-offs that are in line with actual production needs. In the data sensitivity experiment, the ASDINet achieved results that met the production needs under the training of 42 defective pictures. In addition, we also discussed the effect of the ASDINet in actual production, and the test results showed that our proposed network demonstrated excellent performance consistent with the theory in actual production.

Clusters of dead trees are forest fires-prone. To maintain ecological balance and realize its protection, timely detection of dead trees in forest remote sensing images using existing computer vision methods is of great significance. Remote sensing images captured by Unmanned aerial vehicles (UAVs) typically have several issues, e.g., mixed distribution of adjacent but different tree classes, interference of redundant information, and high differences in scales of dead tree clusters, making the detection of dead tree clusters much more challenging. Therefore, based on the Multipath dense composite network (MDCN), an object detection method called LLAM-MDCNet is proposed in this paper. First, a feature extraction network called Multipath dense composite network is designed. The network’s multipath structure can substantially increase the extraction of underlying and semantic features to enhance its extraction capability for rich-information regions. Following that, in the row, column, and diagonal directions, the Longitude Latitude Attention Mechanism (LLAM) is presented and incorporated into the feature extraction network. The multi-directional LLAM facilitates the suppression of irrelevant and redundant information and improves the representation of high-level semantic feature information. Lastly, an AugFPN is employed for down-sampling, yielding a more comprehensive representation of image features with the combination of low-level texture features and high-level semantic information. Consequently, the network’s detection effect for dead tree cluster targets with high-scale differences is improved. Furthermore, we make the collected high-quality aerial dead tree cluster dataset containing 19,517 images shot by drones publicly available for other researchers to improve the work in this paper. Our proposed method achieved 87.25% mAP with an FPS of 66 on our dataset, demonstrating the effectiveness of the LLAM-MDCNet for detecting dead tree cluster targets in forest remote sensing images.

The target detection of smoke through remote sensing images obtained by means of unmanned aerial vehicles (UAVs) can be effective for monitoring early forest fires. However, smoke targets in UAV images are often small and difficult to detect accurately. In this paper, we use YOLOX-L as a baseline and propose a forest smoke detection network based on the parallel spatial domain attention mechanism and a small-scale transformer feature pyramid network (PDAM–STPNNet). First, to enhance the proportion of small forest fire smoke targets in the dataset, we use component stitching data enhancement to generate small forest fire smoke target images in a scaled collage. Then, to fully extract the texture features of smoke, we propose a parallel spatial domain attention mechanism (PDAM) to consider the local and global textures of smoke with symmetry. Finally, we propose a small-scale transformer feature pyramid network (STPN), which uses the transformer encoder to replace all CSP_2 blocks in turn on top of YOLOX-L’s FPN, effectively improving the model’s ability to extract small-target smoke. We validated the effectiveness of our model with recourse to a home-made dataset, the Wildfire Observers and Smoke Recognition Homepage, and the Bowfire dataset. The experiments show that our method has a better detection capability than previous methods.