Fig. 9: Visualization of the deep learning-driven mapping workflow. | Nature Communications

Fig. 9: Visualization of the deep learning-driven mapping workflow.

From: Streaks on martian slopes are dry

Fig. 9

A MRO CTX images are cropped into small patches and fed into the CNN, which (B) outputs predictions of slope streak location, size, and confidence score (true positives, TP, white boxes). The detector can miss slope streaks (false negatives, FN, yellow) or can make incorrect predictions (false positives, FP, red); note how larger streaks feature larger bounding boxes. C Testset performance of the detector is visualized on the left, indicating recall, precision, and AP, as well as the score cut-off of 0.8 used for this study. Image ID B22_018396_1890_XI_09N320W. Raw image credits to NASA/MSSS/ASU.

Back to article page