CN-122024007-A - Small intestine capsule endoscope target detection method with improved YOLOv12

CN122024007ACN 122024007 ACN122024007 ACN 122024007ACN-122024007-A

Abstract

The invention relates to the technical field of image processing, in particular to an improved YOLOv small intestine capsule endoscope target detection method which comprises the steps of collecting small intestine capsule endoscope images, constructing an improved YOLOv model, replacing an A2C2f module in a Backbone network with an A2C2f-MCA module, and replacing all A2C2f modules in a Neck network with an A2C2f-MCA module. The invention solves the problems of weak feature extraction capability and easy loss of details when YOLOv networks process small intestine capsule endoscope targets.

Inventors

YE SHIREN
ZHANG ZETONG
LI LIANGJING
MA HAIPENG
ZHU JIAQUN

Assignees

常州大学

Dates

Publication Date: 20260512
Application Date: 20260121

Claims (8)

1. An improved YOLOv small intestine capsule endoscope target detection method is characterized by comprising the following steps: Step one, collecting an endoscope image of a small intestine capsule; And step two, constructing an improved YOLOv model, replacing an A2C2f module in the backhaul network with an A2C2f-MCA module, and replacing all A2C2f modules in the Neck network with an A2C2f-MCA module.
2. The method for detecting the target of the small intestine capsule endoscope of the improved YOLOv as set forth in claim 1, wherein the A2C2f-MCA module comprises the steps that the characteristic diagram X in is spliced with X in after passing through a first Conv, an ABlock and a second Conv in sequence, and then is input into the MCA module.
3. The improved YOLOv capsule endoscopic target detection method of small intestine according to claim 1, wherein the improved YOLOv12 model further comprises: the upsampling module in Neck network is replaced with a WFE module.
4. The improved YOLOv capsule endoscopic target detection method of small intestine according to claim 1, wherein the image is aligned, scaled and normalized.
5. The improved YOLOv capsule endoscopic target detection method of claim 1, wherein the image is subjected to random hue enhancement, saturation enhancement, brightness enhancement, horizontal flipping, mosiac and Mixup transformations.
6. The improved YOLOv capsule endoscopic target detection method of small intestine according to claim 1, wherein recall and average accuracy mean evaluation is performed on the improved YOLOv model.
7. The small intestine capsule endoscope object detection system of improvement YOLOv12, characterized by comprising a memory for storing instructions executable by a processor, and a processor for executing instructions to implement the small intestine capsule endoscope object detection method of improvement YOLOv according to any of claims 1-6.
8. A computer readable medium storing computer program code, wherein the computer program code when executed by a processor implements the improved method of small intestine capsule endoscope object detection of YOLOv as defined in any one of claims 1-6.

Description

Small intestine capsule endoscope target detection method with improved YOLOv12 Technical Field The invention relates to the technical field of image processing, in particular to an improved YOLOv small intestine capsule endoscope target detection method. Background The capsule endoscope image has the characteristics of large illumination change, obvious focus scale difference, weak texture, high image noise, complex scene and the like, and the traditional convolution detection method is difficult to simultaneously consider local detail and global semantics, and is easy to miss a small-size focus. The patent with the publication number of CN115908224A discloses a main network comprising a plurality of convolution layers and a plurality of RFB modules, wherein the convolution layers and the RFB modules are sequentially connected according to a preset sequence, firstly, the model has higher calculation complexity, a complex characteristic pyramid network and the RFB modules are introduced to increase calculation amount, secondly, the model is improved based on a VGG16 network, the parameter amount is obviously increased, the model deployment is possibly difficult, and finally, the network model has the problem of insufficient generalization capability. YOLOv12 has remarkable improvement on feature expression, multi-scale characterization, detection efficiency and generalization capability, but the problem of insufficient feature utilization and detail information loss still exists when YOLOv12 is directly applied to a capsule endoscope image. Disclosure of Invention Aiming at the defects of the existing method, the invention solves the problems of weak feature extraction capability and easy loss of details when YOLOv networks process small intestine capsule endoscope targets. The technical scheme adopted by the invention is that the method for detecting the target of the small intestine capsule endoscope by improving YOLOv < 12 > comprises the following steps: Step one, collecting an endoscope image of a small intestine capsule; as a preferred embodiment of the present invention, the image is subjected to alignment, scaling and normalization processing. As a preferred embodiment of the present invention, the image is subjected to random hue enhancement, saturation enhancement, brightness enhancement, horizontal flipping, mosiac, and Mixup transforms. Step two, an improved YOLOv model is constructed, and an A2C2f module in a backhaul network is replaced by an A2C2f-MCA module, and all A2C2f modules in a Neck network are replaced by an A2C2f-MCA module; as a preferred embodiment of the invention, the A2C2f-MCA module comprises a characteristic diagram X in which is spliced with X in after passing through a first Conv, an ABlock and a second Conv in sequence and then is input into the MCA module. As a preferred embodiment of the present invention, the improvement YOLOv model further includes replacing the upsampling module in the Neck network with a WFE module. As a preferred embodiment of the invention, recall and average accuracy mean evaluation was performed on the modified YOLOv model. As a preferred embodiment of the present invention, the small intestine capsule endoscope object detection system of improvement YOLOv, comprises a memory for storing instructions executable by a processor, and a processor for executing the instructions to implement the small intestine capsule endoscope object detection method of improvement YOLOv. As a preferred embodiment of the present invention, a computer readable medium storing computer program code which, when executed by a processor, implements the improved YOLOv small intestine capsule endoscope object detection method. The invention has the beneficial effects that: 1. By introducing a lightweight MCA module into the A2C2f architecture of YOLOv, the model can more effectively excavate key semantics from the multi-scale fusion characteristics, and the detection capability of small-size and low-contrast lesions is improved; 2. The low-frequency characteristic and the high-frequency characteristic are respectively processed through wavelet transformation, so that the problem of frequency aliasing is effectively avoided, details in images can be better reserved and recovered, and especially in small intestine capsule endoscope images, the detection precision of a tiny target can be improved. After the traditional up-sampling operation is replaced, the network can better utilize the characteristic information from the encoder, the robustness and the detection precision of the model are improved, and the target detection performance in medical image analysis can be effectively improved. Drawings FIG. 1 is a diagram of a small intestine capsule endoscope object detection model of the improvement YOLOv of the present invention; FIG. 2 is a schematic diagram of an A2C2f module of the present invention; FIG. 3 is a graph showing the detection effect of the improvement YOLOv of