Research on Lizard Target Detection based on YOLOv5s-SSE

In a complex natural environment, the traditional lizard detection method based on image processing is easily affected by the environment, resulting in high false detection and missed detection rates, which has a certain impact on the accurate monitoring and behavioral research of lizards.Therefore, designing an efficient lizard species detection model is convenient for more detailed detection of the distribution of local lizard populations and further study of lizard behavior patterns.To meet the above requirements, this paper proposes a lizard detection model based on YOLOv5s-SSE.This model integrates the Switchable Atrous Convolution (SAC) into the C3 module of the baseline model and combines it into the backbone feature extraction network.It can expand the receptive field and improve the multi-scale feature extraction capability without increasing the convolution kernel; Shuffle Attention mechanism (SA) is adopted to construct the channel attention mechanism and the spatial attention mechanism through grouping, and the information exchange between different groups is strengthened through channel shuffling, which not only better improves the expressive ability of the model, but also takes into account the advantage of lightweight; The EIoU loss function is used to measure the difference between the real box and the predicted box in multiple aspects to strengthen the focus on the real difference between the length and width of the bounding box and its confidence, thereby improving the positioning accuracy of the target bounding box and helping to capture small-scale lizard image information more accurately.This study takes eight species of lizards, including the Australian magic lizard and the red-eyed hawk lizard, as the main research objects, and constructs the LDD (Lizard Detection Dataset) dataset containing 8 types of lizards.Compared with the baseline model, the lizard detection model proposed in this study improves [email protected]:0.95,Precision, and Recall by 1.8%, 0.8%, and 2.2%, respectively.The lizard detection model proposed in this paper has good detection performance in complex environments, reduces the probability of false detection of the detection model, and realizes efficient detection and identification of lizard species.