Dim and Small Target Detection Based on Local Energy Aggregation Degree of Sequence Images

Table 1

Pseudocode of dim small target detection for local energy convergence degree of sequence images.

---

Initialization: let k be transformed three times, k = 3, 4, 5; L × L is the size of target block, L × L = 3 × 3; M is the cumulative frame number, M = 3; Th is the threshold of gradient difference, Th = 5; r is the size of local neighborhood, r = 2; N is the number of frames in the continuous time domain, N = 5; Th₁ is threshold of the total number of times the target moves, Th₁ = 3; Th₂ is the average grayscale threshold of candidate block, Th₂ = 10; Th₃ and Th₄ are the lower and upper bounds of candidate block area, respectively, Th₃ = 4, Th₄ = 9.

---

Input: sequence original images

(1) Sequential difference images are obtained by using an anisotropic gradient background modeling method combined with spatial and temporal information, as shown in equations (1)–(3).

(2) The multidirectional gradient difference of the neighborhood block is constructed for the difference image. For details, see equation (4). According to the gradient difference, the gradient maximum value is obtained by transforming the step of k, and then the binary image is segmented and extracted.

(3) After obtaining the sequence binary images, the pipeline filtering algorithm of the literature [22] is used to obtain the centroid coordinate position (x, y) of the real target.

(4) Centering on the target centroid coordinates (x, y), find the final target point in the local neighborhood (2r + 1, 2r + 1) through equations (5) and (6).

---

Output: Sequence detection results

---