Improved particle filter algorithm combined with culture algorithm for collision Caenorhabditis elegans tracking

Abstract

In order to address the issue of tracking errors of collision Caenorhabditis elegans, this research proposes an improved particle filter tracking method integrated with cultural algorithm. The particle filter algorithm is enhanced through the integration of the sine cosine algorithm, thereby facilitating uninterrupted tracking of the target C. elegans. Furthermore, the cultural algorithm is employed to facilitate recognition of the target C. elegans following a collision. In addition, this method integrates the concepts of down-sample and marking to reduce the average processing time of the image. Ultimately, the experiment was conducted on two strains of C. elegans of six ages. The experimental results demonstrate that the proposed method can accurately identify the target worm in the post-collision stage. The proposed method has the potential to be utilized in the field of worm tracking, offering a novel method into the acquisition of collision C. elegans behavior.

Keywords: Collision C. Elegans tracking; Cultural algorithm; Particle filter; Sine cosine algorithm; Visual object tracking.

Fig. 1

Diagram of typical C. elegans tracking methods and their classification. Among them, section A is model-based worm tracking methods, section B is high-throughput worm tracking methods, section C is an inter-trajectory worm tracking method, and section D is end-to-end tracking methods.

Fig. 2

Flow chart of particle filter tracking C. elegans. Among them, blue area indicates the background, white area indicates the worm blob, the yellow rectangle indicates the bounding box of the worm, the red and gray cross markers indicate the high-weight and low-weight particles, and the green marker indicates the centroid of the worm.

Fig. 3

Flow chart of SCA applied on particles. Among them, the green circle is the target point, the red cross is the particle point, the white area is the area close to the target point, and the yellow area is the area away from the target point. The particle searches for the target point through sine or cosine motion.

Fig. 4

Flow chart of the cultural algorithm for particle filter. The yellow area is the particle space, and the blue area is the belief space. New particles are constantly generated in the particle space through the Select and Objective functions. The characteristics of the particles are accepted by the belief space to form situational knowledge and normative knowledge. This knowledge is used to influence particle generation and determine the correct worm.

Fig. 5

Flowchart of the proposed collision worm tracking method.

Fig. 6

The physical images of the proposed collision worm tracking method. (a) Manually mark the target point. (b) Particles are generated uniformly and updated in the bounding box of worm. (c) After the SCA and resampling stages, the particles aggregate in the worm body. (d) In the pre-collision stage, the particles continuously tracking worm. (e) In the in-collision stage, the particles tracking collision worm area. (f) The particles continuously tracking collision worm. (g) In the post-collision phase, the two worms are tracked separately. (h) Correct worm is determined based on the situational knowledge of the cultural algorithm. Among them, the blue cross represents the target point, the green rectangle is the bounding box of the target worm, and the red dot is the particle. The yellow and blue rectangles are the bounding box of the worms that have not been judged, and the yellow and blue dots are the corresponding particles. The scale bar in (a), (b), (c), (e), (g) and (h) is 2 mm, and the scale bar in (d) and (f) is 3 mm.

Fig. 7

Experimental result figures of the proposed collision worm tracking method. The worm strains are N2 wild type and RB1579, and the ages of worm are L1, L2, L3, L4, Young adult and D1. The green rectangle is the bounding box of the target worm, the red dot is the corresponding particle, and the orange rectangle is the bounding box of the colliding worm. The scale bar of the first three columns is 0.4 mm, and the scale bar of the last three columns is 1 mm.

Fig. 8

The average processing time of collision fragments by different strains. (a) RB1579. (b) N2 wild type. The mean value is between 0.02 and 0.04 s, the maximum value is 0.09 s, and the standard deviation of RB1579 is 0.0136, 0.0077, 0.0109, 0.0159, 0.08 10, and 0.0090, and the standard deviation of N2 wild type is 0.0105, 0.0157, 0.0114, 0.0092, 0.0150, and 0.0139.

Fig. 9

Average error of collision fragments by different strains. (a), (c), (e), (g), (i), (k) The physical images of the N2 strain of L1, L2, L3, L4, Young adult, and D1. (b), (d), (f), (h), (j), (l) The physical images of the RB1579 strain of L1, L2, L3, L4, Young adult, and D1. (m) Average error figure of N2 strain. (n) Average error figure of RB1579 strain. In figures (a) to (l), the green dots represent the expert path, the blue dots represent the proposed method path, and the magenta dots represent the traditional particle filter path. In Figures (m) and (n), the mean is 13 to 15 pixels, the maximum is 20 pixels, and the standard deviation of N2 is 1.5246, 1.4010, 1.5559, 1.4379, 1.65 30, and 1.5367, and the standard deviation for RB1579 is 1.2464, 1.347, 1.2531, 1.2337, 1.2950, and 1.2732. The scale bar in figures (a)–(l) is 1 mm.

Fig. 10

The F-measure of collision fragments by different strains. (a) RB1579. (b) N2 wild type. The median value is between 0.8 and 0.9, the maximum value is 0.98, and the standard deviation of RB1579 is 0.0482, 0.0436, 0.0460, 0.0465, 0.0538, and 0.0487, and the standard deviation of N2 is 0.0467, 0.0422, 0.0433, 0.0430, 0.0452, and 0.0518.