Algorithm 1: Steps for the presented method.

INPUT:

TrD, Ann

OUTPUT:

Localized RoI, EfficientDet, Classified glaucoma diseased portion

TrD—training data.

Ann—Position of the glaucomatous region in suspected images.

Localized RoI—Glaucomatous area in output.

EfficientDet—EfficientNet-B0 based EfficientDet network.

Classified glaucoma diseases portion—Class of identified suspected region.

imageSize ← [x y]

Bbox calculation

µ ← AnchorsCalculation (TrD, Ann)

EfficientDet—Model

EfficientDet ← EfficientNet-B0-Based EfficientDet (imageSize, µ)

[dr dt] ← Splitting database in the training and testing set

The training module of glaucoma recognition

For each sample s in → dr

ExtractEfficientNet-B0-keypoints → ds

Perform features Fusion (ds) → Fs

End

Training EfficientDet on Fs, and compute processing time t_Edet

η_Edet ← DetermineDiseasedPortion(Fs)

Ap_ Edet ← Evaluate_AP (EfficientNet-B0, η_ Edet)

For each image S in → dt

(a) Calculate key points via trained network € → βI

(b) [Bbox, localization_score, class] ← Predict (βI)

(c) Output sample together with Bbox, class

(d) η ← [η Bbox]

End For

Ap_€ ← Evaluate model € employing η

Output_class ← EfficientDet (Ap_€).