Abstract: In order to solve the problems of complex detection scene, low detection accuracy and high computational complexity of tomato leaf pests and diseases in natural environment, a deep learning algorithm SLP-YOLOv7-tiny was proposed. Firstly, part of the 3×3 convolution Conv2D Convolution in the backbone feature extraction network is changed to distributed migration convolution DSConv2D (2D Depthwise Separable Convolution) to reduce the computing load of the network and speed it up. Less memory usage; Secondly, the parameter-free attention module (SimAM) was integrated into the backbone feature extraction network to enhance the model's ability to effectively extract and integrate features of pests and diseases. Finally, the original YOLOv7-tiny CIOU loss function is replaced with Focal EIOU loss function to accelerate the model convergence and reduce the loss value. The test results show that The overall identification accuracy, recall rate, average accuracy mAP_0.5 (the average accuracy when the IOU threshold is 0.5), and mAP_{0.5 ~ 0.95} (the average accuracy when the IOU threshold between 0.5 and 0.95) of SLP-YOLOv7-tiny model are 95.9%, 94.6%, and 98%, respectively Compared with YOLOv7-tiny before the improvement, they respectively increase by 14.7, 29.2, 20.2 and 30 percentage points. Meanwhile, the calculation amount decreases by 62.6% and the detection speed increases by 13.2%. Compared with YOLOv5n, YOLOv5s, YOLOv5m, YOLOv7, Yolov7-Tiny, Faster-RCNN and SSD target detection models, mAP_0.5 improved by 2.0, 1.6, 2.0, 2.2, 20.2, 6.1 and 5.3 percentage points respectively. The computing capacity was only 31.5%, 10.6%, 4.9%, 4.3% and 3.8% of YOLOv5s, YOLOv5m, YOLOv7, father-RCNN and SSD. The results show that SLP-YOLOv7-tiny can accurately and quickly detect tomato leaf diseases and pests, and the model is small, which can provide certain technical support for the development of rapid and accurate detection of tomato leaf diseases and pests.SLP-YOLOv7-tiny can be expected to accurately and rapidly detect the tomato leaf diseases and insect pests. The small model was more conducive to the migration application. A comparison was also performed on the detection performance of different attention mechanisms. The SimAM attention mechanism shared the better screening of effective feature information on the YOLOv7-tiny model. The disease spots were identified to effectively improve the accuracy of the model, compared with the SENet and CBAM attention mechanism. The ablation test was carried out to verified the added module. The high detection accuracy and computational efficiency were suitable for the deployment in the natural environment with the limited computing resources, such as mobile terminals. The visualization of model detection show that the SLP-YOLOv7-tiny model can be used to learn and distinguish the fine features of tomato leaf disease spots. The detection accuracy was also better than that of the YOLOv7-tiny model. Moreover, the SLP-YOLOv7-tiny model can be used to accurately identify the diseases and pests of diseased tomato leaves in the natural shooting environment. The high detection accuracy can be applied for the multiple diseases and pests. The better identification was also obtained for the small features in the similar diseases and pests. The finding can provide the technical support for the rapid and accurate detection of tomato leaf diseases and insect pests.