Food safety has always been a focus of social concern. However, the presence of mosquitoes and other insects during the food packaging and printing process poses a serious challenge to food safety. Aiming at the current situation of manual screening for mosquitoes and other insects during food packaging quality inspection, and account of the small size of insect targets and the complexity of their backgrounds, a fully automatic MNTH-YOLOv8 detection method based on deep learning was proposed. this method was based on the powerful object detection capability of YOLOv8, combined with channel-wise partial convolution modules and SimAM attention mechanism, with CIoU and normalized Wasserstein distance as the localization regression loss function. Experimental results demonstrated significant advantages of the proposed method in real datasets. It not only effectively improved the detection accuracy of small insect targets but also significantly reduced the parameter count while maintaining detection speed, indicating its great prospect in the application of real-time detection of mosquitoes and other insects in food packaging.