Contextualizing CNN attribution maps using predefined image-derived descriptors in medicinal plant seed classification

<title>Abstract</title> <p>Deep learning classifiers can achieve high predictive performance in medicinal plant seed image classification, but prediction accuracy alone does not clarify which visual properties are linked to model responses. This issue is particularly relevant for visually similar medicinal plant seeds, related taxa, and potential substitutes that may share overlapping morphology. Here, we analyzed RGB images from five medicinal plant seed classes by comparing CNN attribution maps with predefined image-derived descriptor maps within the seed foreground. ResNet50, EfficientNet-B0, and ConvNeXt-Small showed near-ceiling classification performance under controlled imaging and preprocessing conditions, and zero-baseline absolute Integrated Gradients (IG) maps from ConvNeXt-Small were used as the primary attribution reference. Among the tested descriptor maps, LAB L, Brightness, and fast Fourier transform (FFT) low-pass showed the highest positive foreground-restricted spatial associations with IG attribution maps. Descriptor summary feature classification further showed that foreground-level descriptor statistics contained class-discriminative visual information. The analysis organizes CNN attribution patterns within reproducible image-derived visual contexts in medicinal plant seed image classification.</p>