Pathological Myopia Image Recognition Strategy Based on Data Augmentation and Model Fusion

Figure 4

DA effect (original dataset image_x: different images in the original dataset: (r), (s), and (t) were mutated from (q)). (a) Original dataset image₁. (b) Randomly change direction. (c) Original dataset image₂. (d) Randomly add Gaussian noise. (e) Original dataset image₃. (f) random color. (g) Original dataset image₄. (h) Random stretching. (i) Original dataset image₅. (j) Randomly adjust the sharpness. (k) Original dataset image₆. (l) Randomly flip, adjust colors, and add Gaussian noise. (m) Original dataset image₇. (n) Randomly flip, adjust colors, and add Gaussian noise. (o) Original dataset image₈. (p) Random stretching and cropping. (q) Original dataset image₉. (r) Random cropping with 0–50 pixels around, 50% probability horizontal flip, and Gaussian blur (sigma = 0 to 3.0). (s) Original dataset image. (t) Randomly stack all operations.

(a)

(b)

(c)

(d)

(e)

(f)

(g)

(h)

(i)

(j)

(k)

(l)

(m)

(n)

(o)

(p)

(q)

(r)

(s)

(t)