With the impact of the COVID‐19 epidemic, the demand for masked face recognition technology has increased. In the process of masked face recognition, some problems such as less feature information and poor robustness to the environment are obvious. The current masked face recognition model is not quantified enough for feature extraction, there are large errors for faces with high similarity, and the categories cannot be clustered during the detection process, resulting in poor classification of masks, which cannot be well adapted to changes in multiple environments. To solve current problems, this paper designs a new masked face recognition model, taking improved Single Shot Multibox Detector (SSD) model as a face detector, and replaces the input layer VGG16 of SSD with Deep Residual Network (ResNet) to increase the receptive field. In order to better adapt to the network, we adjust the convolution kernel size of ResNet. In addition, we fine‐tune the Xception network by designing a new fully connected layer, and reduce the training cycle. The weights of the three input samples including anchor, positive and negative are shared and clustered together with triplet network to improve recognition accuracy. Meanwhile, this paper adjusts alpha parameter in triplet loss. A higher value of alpha can improve the accuracy of model recognition. We further adopt a small trick to classify and predict face feature vectors using multi‐layer perceptron (MLP), and a total of 60 neural nodes are set in the three neural layers of MLP to get higher classification accuracy. Moreover, three datasets of MFDD, RMFRD and SMFRD are fused to obtain high‐quality images in different scenes, and we also add data augmentation and face alignment methods for processing, effectively reducing the interference of the external environment in the process of model recognition. According to the experimental results, the accuracy of masked face recognition reaches 98.3%, it achieves better results compared with other mainstream models. In addition, the hyper‐parameters tuning experiment is carried out to improve the utilization of computing resources, which shows better results than the indicators of different networks.

<abstract> <p>Red imported fire ants (RIFA) are an alien invasive pest that can cause serious ecosystem damage. Timely detection, location and elimination of RIFA nests can further control the spread of RIFA. In order to accurately locate the RIFA nests, this paper proposes an improved deep learning method of YOLOv4. The specific methods were as follows: 1) We improved GhostBottleNeck (GBN) and replaced the original CSP block of YOLOv4, so as to compress the network scale and reduce the consumption of computing resources. 2) An Efficient Channel Attention (ECA) mechanism was introduced into GBN to enhance the feature extraction ability of the model. 3) We used Equalized Focal Loss to reduce the loss value of background noise. 4) We increased and improved the upsampling operation of YOLOv4 to enhance the understanding of multi-layer semantic features to the whole network. 5) CutMix was added in the model training process to improve the model's ability to identify occluded objects. The parameters of improved YOLOv4 were greatly reduced, and the abilities to locate and extract edge features were enhanced. Meanwhile, we used an unmanned aerial vehicle (UAV) to collect images of RIFA nests with different heights and scenes, and we made the RIFA nests (RIFAN) airspace dataset. On the RIFAN dataset, through qualitative analysis of the evaluation indicators, mean average precision (MAP) of the improved YOLOv4 model reaches 99.26%, which is 5.9% higher than the original algorithm. Moreover, compared with Faster R-CNN, SSD and other algorithms, improved YOLOv4 has achieved excellent results. Finally, we transplanted the model to the embedded device Raspberry Pi 4B and assembled it on the UAV, using the model's lightweight and high-efficiency features to achieve flexible and fast flight detection of RIFA nests.</p> </abstract>