Road crack intelligent detection system based on multiple tasks

Abstract

The invention discloses a road crack intelligent detection system based on multi-task optimization. The current domestic road crack detection field mainly depends on manual inspection or single-target automation equipment, and the global optimality of a detection result is difficult to guarantee by the existing method. Aiming at the limitation, a road crack detection problem is modeled into a constraint multi-objective optimization problem based on a multi-task framework, and an intelligent detection system is designed according to the constraint multi-objective optimization problem. According to the system, minimization of comprehensive detection cost and maximization of effective crack detection amount are taken as core optimization targets, and key conditions such as detection equipment endurance constraint, task timeliness requirements, recognition precision threshold and road network coverage rate are strictly met. The system innovatively constructs a collaborative optimization framework comprising a main task and two auxiliary tasks. A high-robustness resource allocation scheme can be generated based on the intelligent optimization process of the mathematical model, and optimization of the overall performance of the system is achieved.

Description

Technical Field

[0001] The present invention relates to the field of intelligent optimization. Based on the actual needs of road inspection, a multi-task road crack detection system is designed. To address the challenges of strong task coupling, multiple constraint conflicts, and high real-time requirements in road infrastructure inspection, a three-population dynamic collaborative optimization mechanism is constructed. While strictly meeting the detection equipment's endurance, task timeliness, and recognition accuracy thresholds, this system simultaneously optimizes crack identification accuracy, road network coverage, and resource consumption costs, providing highly robust intelligent inspection decision support for road maintenance departments. Background Art

[0002] Currently, crack detection in road inspection primarily relies on manual inspections or single-target automated equipment, which presents significant limitations. Manual methods are highly dependent on the experience of maintenance personnel, resulting in low detection efficiency on large-scale road networks, significant missed detections, and difficulty quantifying the risk of disease spread. Existing automated systems often focus on a single optimization objective, ignoring the multi-objective coupling between crack identification accuracy, task coverage, and resource consumption, resulting in insufficient robustness in detection solutions. Traditional evolutionary algorithms are prone to falling into local optimal solutions when dealing with multiple constraints, such as equipment endurance, task time window conflicts, and network connectivity, making it difficult to ensure task completion in complex scenarios. With the acceleration of urbanization and the continued expansion of road infrastructure, there is an urgent need for intelligent detection technologies that can simultaneously meet constraints and optimize multi-objective conflicts.

[0003] Based on detailed research with road inspection companies, we developed a multi-task road crack detection system. Compared to manual inspections or single-target automated approaches, this system can achieve a more optimal allocation plan in the shortest possible time, ultimately maximizing profits. Preliminary calculations using data from a road inspection company indicate a profit increase of at least 9% compared to manual inspections or single-target automated approaches. Driven by the global trend toward intelligent infrastructure, high-precision road crack detection technology has become a rigid requirement for smart transportation development, and its implementation and large-scale application present a significant market opportunity. Intelligent road crack detection involves hardware resources provided by edge computing platforms, multi-source road surface features acquired by image acquisition devices, and real-time environmental parameters. The system primarily considers minimizing total cost, which includes comprehensive costs such as accuracy loss and latency penalties, while also addressing hardware resource constraints, real-time requirements, and specific detection scenarios. Based on specific needs, the system intelligently optimizes the allocation plan to achieve the optimal solution, ultimately maximizing overall profits. Summary of the Invention

[0004] The implementation of the present invention is mainly divided into two parts: actual demand modeling and algorithm optimization.

[0005] 1. Actual demand modeling

[0006] The intelligent road crack detection problem can be modeled as a constrained multi-objective resource optimization problem. While meeting hardware resource constraints and real-time requirements, the system simultaneously optimizes two core objectives: minimizing overall detection costs and maximizing the number of effective crack detections. If customized adjustments are required based on specific requirements, the model can be flexibly modified, such as adding or removing constraints, to meet specific needs.

[0007] 2. Algorithm optimization

[0008] The optimization process for road crack detection is often modeled as a constrained multi-objective optimization problem. The core challenge of this type of problem lies in the interdependence of multiple objective functions (such as detection accuracy, processing speed, and resource consumption): optimizing one objective often negatively impacts others. Therefore, seeking a single "optimal solution" is often infeasible. Instead, the goal is to identify a set of non-dominated (Pareto-optimal) solutions—the Pareto front. These solutions represent the optimal trade-off between objectives that cannot be simultaneously improved under given constraints.

[0009] The introduction of constraints (such as lower limits on detection accuracy, real-time requirements, and upper limits on computing resources) significantly increases the complexity of the problem. This strictly divides the solution space into a feasible region (solutions that satisfy all constraints) and an infeasible region. Optimization algorithms must explore within the feasible region while effectively coordinating multiple conflicting objectives, making the search for a high-quality Pareto front significantly more challenging. Constrained multi-objective evolutionary algorithms are an effective tool for solving such problems. Their core goal is to drive the population toward a Pareto front that simultaneously approaches the optimal values ​​of multiple objectives while strictly satisfying the constraints.

[0010] In recent years, multi-task optimization has provided a novel and effective approach for solving constrained multi-objective problems. Compared to traditional constrained multi-objective evolutionary algorithms, multi-task-based approaches exhibit a unique advantage: they can collaboratively optimize multiple related tasks in parallel (for example, simultaneously optimizing different submodules of crack detection or detection models for different scenarios). This inter-task knowledge transfer and resource sharing mechanism often enables more efficient exploration of the solution space, leading to the identification of more optimal and robust Pareto solutions under complex constraints, making it a promising research direction in this field.

[0011] Currently, the domestic road crack detection field mainly relies on manual inspections or single-target automated design. These methods often make it difficult to guarantee near-optimal or optimal image quality. In contrast, intelligent road crack detection optimization can achieve optimal or near-optimal results in a shorter timeframe, saving time and reducing costs.

[0012] The advantages of the present invention are:

[0013] The HPCHEA algorithm is improved and adopts a multi-task framework. The main advantages of this algorithm are as follows:

[0014] 1. Two auxiliary tasks were constructed, adopting a dynamic population collaboration method. In the early stage of evolution, the main population and auxiliary population 1 used a weak collaboration method, and in the later stage of evolution, the main population and auxiliary population 2 used a strong collaboration method.

[0015] 2. Knowledge transfer is a key part of evolutionary multitasking, and its quality directly affects the final results. Selecting different information for interaction based on different situations effectively improves the diversity of the population, thereby helping to escape from local optimal areas.

[0016] 3. The three populations adopted different evolutionary strategies in the early and late stages of evolution. In the early stage of evolution, the first auxiliary population adopted a constraint relaxation strategy to expand the search range. In the late stage of evolution, the second auxiliary population adopted a feasibility principle strategy to ensure the feasibility of the population.

[0017] Therefore, the improved HPCHEA algorithm is a suitable core solver for the road crack detection optimization problem. By integrating the strengths of multiple evolutionary strategies, this algorithm effectively handles conflicts among multiple objectives and approaches a high-quality, well-distributed Pareto frontier under complex constraints. It is important to emphasize that algorithm selection should be tailored to model requirements. In practical application scenarios, crack detection optimization models built based on different requirements (such as real-time performance, detection accuracy priorities, and hardware resource constraints) often possess unique structural characteristics and constraints. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a schematic diagram of the road crack detection process.

[0019] Figure 2 It is a schematic diagram of the modeling process.

[0020] Figure 3 Schematic diagram of the solution process DETAILED DESCRIPTION

[0021] The present invention designs an intelligent road crack detection system based on multi-task optimization. The core design of the present invention is to construct multiple closely related optimization task models, and use the multi-task optimization framework to realize knowledge transfer and co-evolution, so as to efficiently solve the complex constrained multi-objective optimization problem in road crack detection. Minimizing the comprehensive detection cost and maximizing the effective crack detection amount are set as two optimization goals. At the same time, factors such as hardware resource constraints and real-time constraints are used as constraints to construct a constrained multi-objective optimization model as the main task; a model with relaxed constraints is constructed as auxiliary task one; and a model focusing on different optimization perspectives is constructed as auxiliary task two. Finally, the improved HPCHEA algorithm is used as the solver of the intelligent road crack detection system based on the multi-task model for optimization.

[0022] The detailed steps are as follows:

[0023] Step 1: Enter data

[0024] Convert the data in Excel to a .mat file and read it into the MATLAB buffer.

[0025] Step 2: Problem Modeling

[0026] During the problem modeling phase, we clearly defined the decision variables and objective functions for each task. These decision variables primarily included hardware resources, real-time performance, detection accuracy limits, road network coverage, device endurance, and network connectivity. Our goal was to minimize overall detection costs and maximize the number of cracks detected.

[0027] Through problem modeling, we can clearly describe the decision variables and objective function of the problem. Based on the specific requirements of the current inspection task, we then instantiate the mathematical models of the main task, auxiliary task one, and auxiliary task two defined above, as well as the specific calculation formulas for the objective function and constraint function of each task. This provides a foundation for subsequent optimization solutions.

[0028] Step 3: Solve

[0029] The improved HPCHEA algorithm is used for optimization. After the algorithm terminates, the Pareto optimal solution set is extracted from the final generation of the main population. Each solution represents a specific road crack detection resource allocation and scheduling plan, including:

[0030] 1. The detection algorithm and parameters used for each road segment / image patch;

[0031] 2. The computing resources (number of CPU cores, memory size, etc.) allocated to each detection task and the order or timing of task execution;

[0032] 3. Estimated overall cost, effective crack detection volume, resource consumption, processing time, etc.

[0033] The system outputs the optimal solution to the road maintenance department, guiding the actual inspection vehicles / drones to perform the task, or deploying the corresponding inspection service on the edge computing platform. The system can periodically rerun the optimization process based on new inspection requirements or environmental changes.

Claims

1. A multi-task road crack intelligent detection system, characterized by The detection scenarios are complex and the task coupling is strong. The system includes: weight calculation module, optimization solution module, and intelligent detection scheduling module; The weight calculation module is used to fuse multi-source detection data, build a multi-task coupling weight model for crack detection, and transform the detection area scheduling problem into a constrained multi-objective optimization problem; The optimization solution module is used to solve the constrained multi-objective optimization model and adopts a multi-task hybrid collaborative optimization framework to perform efficient solution space search; The intelligent detection scheduling module is used to generate the optimal detection path strategy, perform robustness analysis and task conflict resolution on the optimized solution set, and output an executable detection plan.

2. The multi-task road crack intelligent detection system according to claim 1, characterized in that: The weight calculation module associates historical road disease data, real-time traffic flow, detection equipment parameters and meteorological information, constructs a multi-objective optimization model that includes crack identification accuracy, disease diffusion risk coefficient, equipment energy consumption cost, detection timeliness weight, and task priority, and outputs the regional detection weight matrix through a feature fusion network.

3. The multi-task road crack intelligent detection system according to claim 1, characterized in that: The optimization solution module adopts a multi-task constraint collaborative optimization framework (improved HPCHEA algorithm) for solution, specifically including: (a) Establish a main population, a first auxiliary population, and a second auxiliary population; the main population strictly adheres to the detection equipment endurance constraints, task time window constraints, and disease identification accuracy thresholds; the first auxiliary population uses a constraint relaxation mechanism to allow the upper limit of equipment endurance to be exceeded in the early stages of evolution to explore the solution space with high objective function values; the second auxiliary population adopts the feasibility priority principle and only retains strongly feasible solutions that meet road network connectivity and task dependencies; (b) A weak collaboration strategy is implemented in the early stages of evolution, where the three populations evolve independently and achieve one-way knowledge transfer through archiving elite solutions. A strong collaboration strategy is initiated in the late stages of evolution, where the high-payoff solutions of the first auxiliary population and the strongly feasible solutions of the second auxiliary population are injected into the main population through the crossover operator, and a constraint repair operator is used to eliminate solution conflicts. (c) Simultaneously optimize the three objectives of crack identification accuracy, detection task coverage, and total equipment energy consumption cost; generate a Pareto optimal solution set through non-dominated sorting to support the decision-making of the intelligent detection scheduling module.

4. The multi-task road crack intelligent detection system according to claim 1, characterized in that: The intelligent detection scheduling module integrates the optimized solution set with real-time road condition data to generate an anti-interference detection strategy. It coordinates the movement trajectories of multiple detection devices through a spatiotemporal conflict resolution algorithm, maximizing the road network disease coverage and minimizing inspection resource consumption while meeting the detection accuracy threshold, equipment endurance limit, and task deadline.

Images