LAUSR

Automatic pest detection and recognition using computer vision techniques are a hot topic in modern intelligent agriculture but suffer from a serious challenge: difficulty distinguishing the targets of similar pests in 2D images. The appearance-similarity problem could be summarized into two aspects: texture similarity and scale similarity. In this paper, we re-consider the pest similarity problem and state a new task for the specific agricultural pest detection, namely Appearance Similarity Pest Detection (ASPD) task. Specifically, we propose two novel metrics to define the texture-similarity and scale-similarity problems quantitatively, namely Multi-Texton Histogram (MTH) and Object Relative Size (ORS). Following the new definition of ASPD, we build a task-specific dataset named PestNet-AS that is collected and re-annotated from PestNet dataset and also present a corresponding method ASP-Det. In detail, our ASP-Det is designed to solve the texture-similarity by proposing a Pairwise Self-Attention (PSA) mechanism and Non-Local Modules to construct a domain adaptive balanced feature module that could provide high-quality feature descriptors for accurate pest classification. We also present a Skip-Calibrated Convolution (SCC) module that can balance the scale variation among the pest objects and re-calibrate the feature maps into the sizing equivalent of pests. Finally, ASP-Det integrates the PSA-Non Local and SCC modules into a one-stage anchor-free detection framework with a center-ness localization mechanism. Experiments on PestNet-AS show that our ASP-Det could serve as a strong baseline for the ASPD task.