Martian Pitted Cone Identification Based on EMHA-BiFPN-SSD

The identification of pitted cone is of great significance in Mars exploration. The traditional methods mainly rely on manual visual interpretation, which is highly inefficient, while recent deep-learning techniques still require human intervention for accurate identification. To solve these problems, we propose a new object detection algorithm, EMHA-BiFPN-SSD (Efficient Multi-Head Attention-Bidirectional Feature Pyramid-SSD). Based on the SSD, EMHA-BiFPN-SSD introduces an efficient multi-head attention mechanism and replaces the traditional Feature Pyramid Network (FPN) with a bidirectional feature pyramid. These improvements enhance the recognition and extraction ability of multi-scale cones and boost the accuracy of detection. Experimental results demonstrate that our model achieves an accuracy of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{8 2. 0 7 \%}$</tex> and an F1-score of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{8 1. 0 2 \%}$</tex> on CTX images, outperforming other models.