Since the AlphaFold protein database provides predicted structures for many UniProt proteins, you can try the following with iCn3D. First, open a UniProt ID with the menu "File > Retrieve by ID > AlphaFold UniProt ID", e.g., https://www.ncbi.nlm.nih.gov/Structure/icn3d/full.html?afid=A0A061AD48. Then open the detailed sequence view "Analysis > Seq. & Annotations : Details" to see the secondary structures. This whole process can be written in a Node.js script to run in command line. Some examples are at https://github.com/ncbi/icn3d/tree/master/icn3dnode.

Interactively, you can click "Helix Sets" and "Sheet Sets" button in the sequence window to generate sets. Finally export the sets in a text file with the menu "File > Save Files > Selection File".

I think one of the best solutions is to write a (Python) script yourself. SVM and PSIPRED prediction methods show higher Q3, you can base on algorithms of one of these two. Run PSI-BLAST of your fasta files against your DB to extract sequence profiles, then generate scoring matrices, do cross-validation and statistical validaton.