目录
1. EfficientNet
EfficientNet是一种卷积神经网络（CNN）架构，由谷歌的研究人员于2019年推出。它以高效利用计算资源而闻名，同时在各种计算机视觉任务（如图像分类和对象检测）上实现了最先进的性能。
EfficientNet背后的关键思想是通过优化三个维度（深度、宽度和分辨率）以更平衡的方式扩展模型架构。通常，当放大CNN时，研究人员会增加层数（深度）、每层中的滤波器数量（宽度）和输入图像分辨率。然而，这可能会迅速导致计算要求的增加，而性能没有相应的提高。
为了解决这个问题，研究人员引入了一种复合缩放方法，该方法使用复合系数同时缩放网络深度、宽度和分辨率。该系数允许有效的模型缩放，因为它根据计算效率和精度之间的预定义权衡来确定每个维度增加多少。
EfficientNet通过结合使用众所周知的技术来实现高效率，例如深度可分离卷积，它减少了计算次数，以及挤压和激励块，这有助于网络专注于重要特征。此外，它利用神经架构搜索（NAS）算法在给定固定资源约束的情况下自动发现最优架构。
总体而言，EfficientNet在各种图像分类基准测试（如ImageNet）上表现出了卓越的性能，同时与以前最先进的模型相比，需要更少的参数和计算。其高效的设计使其非常适合资源受限的应用程序，如移动设备和边缘计算。
其中，EfficientNet 网络结构如下：
2. EfficientNet 实现的花分类
EfficientNet 实现的model部分代码如下面所示，这里如果采用官方预训练权重的话，会自动导入官方提供的最新版本的权重
if model == 'b0':
net = m.efficientnet_b0(weights=m.EfficientNet_B0_Weights.DEFAULT if weights else False,progress=True)
elif model == 'b1':
net = m.efficientnet_b1(weights=m.EfficientNet_B1_Weights.DEFAULT if weights else False,progress=True)
elif model == 'b2':
net = m.efficientnet_b2(weights=m.EfficientNet_B2_Weights.DEFAULT if weights else False,progress=True)
elif model == 'b3':
net = m.efficientnet_b3(weights=m.EfficientNet_B3_Weights.DEFAULT if weights else False,progress=True)
elif model == 'b4':
net = m.efficientnet_b4(weights=m.EfficientNet_B4_Weights.DEFAULT if weights else False,progress=True)
elif model == 'b5':
net = m.efficientnet_b5(weights=m.EfficientNet_B5_Weights.DEFAULT if weights else False,progress=True)
elif model == 'b6':
net = m.efficientnet_b6(weights=m.EfficientNet_B6_Weights.DEFAULT if weights else False,progress=True)
elif model == 'b7':
net = m.efficientnet_b7(weights=m.EfficientNet_B7_Weights.DEFAULT if weights else False,progress=True)
else:
print('模型选择错误！！')
return None2.1 花数据集
数据集文件如下：
标签如下：这里没找到对应的中文标签....
{
"0": "0",
"1": "1",
"2": "10",
"3": "100",
"4": "101",
"5": "102",
"6": "103",
"7": "11",
"8": "12",
"9": "13",
"10": "14",
"11": "15",
"12": "16",
"13": "17",
"14": "18",
"15": "19",
"16": "2",
"17": "20",
"18": "21",
"19": "22",
"20": "23",
"21": "24",
"22": "25",
"23": "26",
"24": "27",
"25": "28",
"26": "29",
"27": "3",
"28": "30",
"29": "31",
"30": "32",
"31": "33",
"32": "34",
"33": "35",
"34": "36",
"35": "37",
"36": "38",
"37": "39",
"38": "4",
"39": "40",
"40": "41",
"41": "42",
"42": "43",
"43": "44",
"44": "45",
"45": "46",
"46": "47",
"47": "48",
"48": "49",
"49": "5",
"50": "50",
"51": "51",
"52": "52",
"53": "53",
"54": "54",
"55": "55",
"56": "56",
"57": "57",
"58": "58",
"59": "59",
"60": "6",
"61": "60",
"62": "61",
"63": "62",
"64": "63",
"65": "64",
"66": "65",
"67": "66",
"68": "67",
"69": "68",
"70": "69",
"71": "7",
"72": "70",
"73": "71",
"74": "72",
"75": "73",
"76": "74",
"77": "75",
"78": "76",
"79": "77",
"80": "78",
"81": "79",
"82": "8",
"83": "80",
"84": "81",
"85": "82",
"86": "83",
"87": "84",
"88": "85",
"89": "86",
"90": "87",
"91": "88",
"92": "89",
"93": "9",
"94": "90",
"95": "91",
"96": "92",
"97": "93",
"98": "94",
"99": "95",
"100": "96",
"101": "97",
"102": "98",
"103": "99"
}其中，训练集的总数为12750，验证集的总数为3712
2.2 训练脚本
训练的参数如下：
parser.add_argument("--model", default='b0', type=str,help='b0,b1,b2,b3,b4,b5,b6,b7')
parser.add_argument("--pretrained", default=True, type=bool) # 采用官方权重
parser.add_argument("--freeze_layers", default=True, type=bool) # 冻结权重
parser.add_argument("--batch-size", default=32, type=int)
parser.add_argument("--epochs", default=100, type=int)
parser.add_argument("--optim", default='SGD', type=str,help='SGD,Adam,AdamW') # 优化器选择
parser.add_argument('--lr', default=0.01, type=float)
parser.add_argument('--lrf',default=0.0001,type=float) # 最终学习率 = lr * lrf
parser.add_argument('--save_ret', default='runs', type=str) # 保存结果
parser.add_argument('--data_train',default='./data/train',type=str) # 训练集路径
parser.add_argument('--data_val',default='./data/val',type=str) # 测试集路径网络分类的个数不需要指定，摆放好数据集后，代码会根据数据集自动生成！
2.3 训练结果
所有的结果都保存在 save_ret 目录下，这里是 runs ：
weights 下有最好和最后的权重，在训练完成后控制台会打印最好的epoch
这里只展示部分结果：好像过拟合了...
训练日志：
"epoch:99": {
"train info": {
"accuracy": 0.9091764705875222,
"0": {
"Precision": 0.8942,
"Recall": 0.9108,
"Specificity": 0.9977,
"F1 score": 0.9024
},
"1": {
"Precision": 1.0,
"Recall": 0.9615,
"Specificity": 1.0,
"F1 score": 0.9804
},
"10": {
"Precision": 0.8357,
"Recall": 0.8603,
"Specificity": 0.9982,
"F1 score": 0.8478
},
"100": {
"Precision": 0.9091,
"Recall": 0.9677,
"Specificity": 0.9998,
"F1 score": 0.9375
},
"101": {
"Precision": 0.9615,
"Recall": 1.0,
"Specificity": 0.9999,
"F1 score": 0.9804
},
"102": {
"Precision": 0.8049,
"Recall": 0.8359,
"Specificity": 0.9936,
"F1 score": 0.8201
},
"103": {
"Precision": 0.8625,
"Recall": 0.9031,
"Specificity": 0.9911,
"F1 score": 0.8823
},
"11": {
"Precision": 0.9767,
"Recall": 0.9767,
"Specificity": 0.9999,
"F1 score": 0.9767
},
"12": {
"Precision": 0.9783,
"Recall": 0.9783,
"Specificity": 0.9998,
"F1 score": 0.9783
},
"13": {
"Precision": 0.8959,
"Recall": 0.9163,
"Specificity": 0.9978,
"F1 score": 0.906
},
"14": {
"Precision": 0.9386,
"Recall": 0.9427,
"Specificity": 0.9989,
"F1 score": 0.9406
},
"15": {
"Precision": 0.913,
"Recall": 1.0,
"Specificity": 0.9998,
"F1 score": 0.9545
},
"16": {
"Precision": 0.9483,
"Recall": 1.0,
"Specificity": 0.9998,
"F1 score": 0.9735
},
"17": {
"Precision": 0.9792,
"Recall": 0.94,
"Specificity": 0.9999,
"F1 score": 0.9592
},
"18": {
"Precision": 0.9412,
"Recall": 0.8889,
"Specificity": 0.9996,
"F1 score": 0.9143
},
"19": {
"Precision": 0.9615,
"Recall": 0.9615,
"Specificity": 0.9999,
"F1 score": 0.9615
},
"2": {
"Precision": 1.0,
"Recall": 0.8,
"Specificity": 1.0,
"F1 score": 0.8889
},
"20": {
"Precision": 1.0,
"Recall": 1.0,
"Specificity": 1.0,
"F1 score": 1.0
},
"21": {
"Precision": 0.9535,
"Recall": 0.8542,
"Specificity": 0.9997,
"F1 score": 0.9011
},
"22": {
"Precision": 0.9388,
"Recall": 0.9583,
"Specificity": 0.9998,
"F1 score": 0.9484
},
"23": {
"Precision": 1.0,
"Recall": 0.9474,
"Specificity": 1.0,
"F1 score": 0.973
},
"24": {
"Precision": 0.9535,
"Recall": 0.9647,
"Specificity": 0.9997,
"F1 score": 0.9591
},
"25": {
"Precision": 0.9359,
"Recall": 0.8795,
"Specificity": 0.9996,
"F1 score": 0.9068
},
"26": {
"Precision": 1.0,
"Recall": 1.0,
"Specificity": 1.0,
"F1 score": 1.0
},
"27": {
"Precision": 1.0,
"Recall": 0.9706,
"Specificity": 1.0,
"F1 score": 0.9851
},
"28": {
"Precision": 0.9421,
"Recall": 0.958,
"Specificity": 0.9994,
"F1 score": 0.95
},
"29": {
"Precision": 0.8932,
"Recall": 0.844,
"Specificity": 0.9991,
"F1 score": 0.8679
},
"3": {
"Precision": 1.0,
"Recall": 0.5714,
"Specificity": 1.0,
"F1 score": 0.7272
},
"30": {
"Precision": 0.8911,
"Recall": 0.8571,
"Specificity": 0.9991,
"F1 score": 0.8738
},
"31": {
"Precision": 0.875,
"Recall": 0.875,
"Specificity": 0.9998,
"F1 score": 0.875
},
"32": {
"Precision": 1.0,
"Recall": 1.0,
"Specificity": 1.0,
"F1 score": 1.0
},
"33": {
"Precision": 0.9524,
"Recall": 1.0,
"Specificity": 0.9999,
"F1 score": 0.9756
},
"34": {
"Precision": 0.9474,
"Recall": 1.0,
"Specificity": 0.9999,
"F1 score": 0.973
},
"35": {
"Precision": 1.0,
"Recall": 0.9444,
"Specificity": 1.0,
"F1 score": 0.9714
},
"36": {
"Precision": 0.9828,
"Recall": 1.0,
"Specificity": 0.9999,
"F1 score": 0.9913
},
"37": {
"Precision": 0.9286,
"Recall": 1.0,
"Specificity": 0.9998,
"F1 score": 0.963
},
"38": {
"Precision": 0.9412,
"Recall": 0.8421,
"Specificity": 0.9999,
"F1 score": 0.8889
},
"39": {
"Precision": 0.971,
"Recall": 0.9178,
"Specificity": 0.9998,
"F1 score": 0.9437
},
"4": {
"Precision": 0.8583,
"Recall": 0.8791,
"Specificity": 0.9915,
"F1 score": 0.8686
},
"40": {
"Precision": 0.9846,
"Recall": 1.0,
"Specificity": 0.9999,
"F1 score": 0.9922
},
"41": {
"Precision": 0.9348,
"Recall": 0.8958,
"Specificity": 0.9995,
"F1 score": 0.9149
},
"42": {
"Precision": 0.9508,
"Recall": 0.9206,
"Specificity": 0.9998,
"F1 score": 0.9355
},
"43": {
"Precision": 0.9545,
"Recall": 0.9545,
"Specificity": 0.9996,
"F1 score": 0.9545
},
"44": {
"Precision": 0.9474,
"Recall": 1.0,
"Specificity": 0.9999,
"F1 score": 0.973
},
"45": {
"Precision": 0.9349,
"Recall": 0.9186,
"Specificity": 0.9991,
"F1 score": 0.9267
},
"46": {
"Precision": 0.9661,
"Recall": 0.912,
"Specificity": 0.9997,
"F1 score": 0.9383
},
"47": {
"Precision": 0.9219,
"Recall": 0.9502,
"Specificity": 0.9983,
"F1 score": 0.9358
},
"48": {
"Precision": 0.9147,
"Recall": 0.9147,
"Specificity": 0.9971,
"F1 score": 0.9147
},
"49": {
"Precision": 0.9139,
"Recall": 0.9432,
"Specificity": 0.9959,
"F1 score": 0.9283
},
"5": {
"Precision": 0.9524,
"Recall": 0.9195,
"Specificity": 0.9997,
"F1 score": 0.9357
},
"50": {
"Precision": 0.8966,
"Recall": 0.9055,
"Specificity": 0.9983,
"F1 score": 0.901
},
"51": {
"Precision": 0.9208,
"Recall": 0.8857,
"Specificity": 0.9994,
"F1 score": 0.9029
},
"52": {
"Precision": 0.9109,
"Recall": 0.8,
"Specificity": 0.9993,
"F1 score": 0.8519
},
"53": {
"Precision": 0.9531,
"Recall": 0.9283,
"Specificity": 0.9983,
"F1 score": 0.9405
},
"54": {
"Precision": 0.9474,
"Recall": 0.973,
"Specificity": 0.9998,
"F1 score": 0.96
},
"55": {
"Precision": 1.0,
"Recall": 0.9655,
"Specificity": 1.0,
"F1 score": 0.9824
},
"56": {
"Precision": 0.8925,
"Recall": 0.9326,
"Specificity": 0.9992,
"F1 score": 0.9121
},
"57": {
"Precision": 0.9839,
"Recall": 0.9683,
"Specificity": 0.9999,
"F1 score": 0.976
},
"58": {
"Precision": 1.0,
"Recall": 1.0,
"Specificity": 1.0,
"F1 score": 1.0
},
"59": {
"Precision": 1.0,
"Recall": 0.9828,
"Specificity": 1.0,
"F1 score": 0.9913
},
"6": {
"Precision": 0.9474,
"Recall": 1.0,
"Specificity": 0.9999,
"F1 score": 0.973
},
"60": {
"Precision": 1.0,
"Recall": 1.0,
"Specificity": 1.0,
"F1 score": 1.0
},
"61": {
"Precision": 0.9286,
"Recall": 0.8966,
"Specificity": 0.9998,
"F1 score": 0.9123
},
"62": {
"Precision": 0.9462,
"Recall": 0.9462,
"Specificity": 0.9996,
"F1 score": 0.9462
},
"63": {
"Precision": 1.0,
"Recall": 1.0,
"Specificity": 1.0,
"F1 score": 1.0
},
"64": {
"Precision": 1.0,
"Recall": 0.9818,
"Specificity": 1.0,
"F1 score": 0.9908
},
"65": {
"Precision": 1.0,
"Recall": 1.0,
"Specificity": 1.0,
"F1 score": 1.0
},
"66": {
"Precision": 1.0,
"Recall": 0.9048,
"Specificity": 1.0,
"F1 score": 0.95
},
"67": {
"Precision": 0.935,
"Recall": 0.9386,
"Specificity": 0.9957,
"F1 score": 0.9368
},
"68": {
"Precision": 0.81,
"Recall": 0.8692,
"Specificity": 0.9958,
"F1 score": 0.8386
},
"69": {
"Precision": 0.9271,
"Recall": 0.9468,
"Specificity": 0.9994,
"F1 score": 0.9368
},
"7": {
"Precision": 0.9712,
"Recall": 0.9619,
"Specificity": 0.9998,
"F1 score": 0.9665
},
"70": {
"Precision": 0.9619,
"Recall": 0.9712,
"Specificity": 0.9997,
"F1 score": 0.9665
},
"71": {
"Precision": 0.7763,
"Recall": 0.8613,
"Specificity": 0.9973,
"F1 score": 0.8166
},
"72": {
"Precision": 0.9277,
"Recall": 0.9222,
"Specificity": 0.999,
"F1 score": 0.9249
},
"73": {
"Precision": 0.8358,
"Recall": 0.8522,
"Specificity": 0.9937,
"F1 score": 0.8439
},
"74": {
"Precision": 0.871,
"Recall": 0.864,
"Specificity": 0.9987,
"F1 score": 0.8675
},
"75": {
"Precision": 0.9051,
"Recall": 0.8725,
"Specificity": 0.9977,
"F1 score": 0.8885
},
"76": {
"Precision": 1.0,
"Recall": 0.9916,
"Specificity": 1.0,
"F1 score": 0.9958
},
"77": {
"Precision": 0.9225,
"Recall": 0.8561,
"Specificity": 0.9992,
"F1 score": 0.8881
},
"78": {
"Precision": 0.9398,
"Recall": 0.907,
"Specificity": 0.9996,
"F1 score": 0.9231
},
"79": {
"Precision": 0.9576,
"Recall": 0.9576,
"Specificity": 0.9996,
"F1 score": 0.9576
},
"8": {
"Precision": 0.9048,
"Recall": 0.8736,
"Specificity": 0.9994,
"F1 score": 0.8889
},
"80": {
"Precision": 0.9803,
"Recall": 0.9739,
"Specificity": 0.9998,
"F1 score": 0.9771
},
"81": {
"Precision": 0.9032,
"Recall": 0.8317,
"Specificity": 0.9993,
"F1 score": 0.866
},
"82": {
"Precision": 0.9244,
"Recall": 0.8209,
"Specificity": 0.9993,
"F1 score": 0.8696
},
"83": {
"Precision": 0.9151,
"Recall": 0.8661,
"Specificity": 0.9993,
"F1 score": 0.8899
},
"84": {
"Precision": 0.9394,
"Recall": 1.0,
"Specificity": 0.9998,
"F1 score": 0.9688
},
"85": {
"Precision": 0.9062,
"Recall": 1.0,
"Specificity": 0.9998,
"F1 score": 0.9508
},
"86": {
"Precision": 0.7812,
"Recall": 0.8333,
"Specificity": 0.9978,
"F1 score": 0.8064
},
"87": {
"Precision": 0.8725,
"Recall": 0.8904,
"Specificity": 0.9985,
"F1 score": 0.8814
},
"88": {
"Precision": 0.95,
"Recall": 0.9896,
"Specificity": 0.9996,
"F1 score": 0.9694
},
"89": {
"Precision": 0.8776,
"Recall": 0.9348,
"Specificity": 0.9995,
"F1 score": 0.9053
},
"9": {
"Precision": 0.9213,
"Recall": 0.9762,
"Specificity": 0.9994,
"F1 score": 0.948
},
"90": {
"Precision": 0.9505,
"Recall": 0.9057,
"Specificity": 0.9996,
"F1 score": 0.9276
},
"91": {
"Precision": 0.9292,
"Recall": 0.9459,
"Specificity": 0.9994,
"F1 score": 0.9375
},
"92": {
"Precision": 0.9565,
"Recall": 0.9167,
"Specificity": 0.9999,
"F1 score": 0.9362
},
"93": {
"Precision": 0.9773,
"Recall": 0.9281,
"Specificity": 0.9998,
"F1 score": 0.9521
},
"94": {
"Precision": 0.8571,
"Recall": 0.8244,
"Specificity": 0.9986,
"F1 score": 0.8404
},
"95": {
"Precision": 0.864,
"Recall": 0.8504,
"Specificity": 0.9987,
"F1 score": 0.8571
},
"96": {
"Precision": 0.8987,
"Recall": 0.71,
"Specificity": 0.9994,
"F1 score": 0.7933
},
"97": {
"Precision": 0.875,
"Recall": 0.8537,
"Specificity": 0.9996,
"F1 score": 0.8642
},
"98": {
"Precision": 0.9714,
"Recall": 1.0,
"Specificity": 0.9999,
"F1 score": 0.9855
},
"99": {
"Precision": 1.0,
"Recall": 1.0,
"Specificity": 1.0,
"F1 score": 1.0
},
"mean precision": 0.9362067307692311,
"mean recall": 0.9250490384615382,
"mean specificity": 0.9990980769230762,
"mean f1 score": 0.9293653846153844
},
"valid info": {
"accuracy": 0.725484913791149,
"0": {
"Precision": 0.7955,
"Recall": 0.8861,
"Specificity": 0.995,
"F1 score": 0.8384
},
"1": {
"Precision": 1.0,
"Recall": 0.8571,
"Specificity": 1.0,
"F1 score": 0.9231
},
"10": {
"Precision": 0.6,
"Recall": 0.6,
"Specificity": 0.9956,
"F1 score": 0.6
},
"100": {
"Precision": 0.75,
"Recall": 0.6667,
"Specificity": 0.9995,
"F1 score": 0.7059
},
"101": {
"Precision": 1.0,
"Recall": 0.8571,
"Specificity": 1.0,
"F1 score": 0.9231
},
"102": {
"Precision": 0.5038,
"Recall": 0.5877,
"Specificity": 0.9817,
"F1 score": 0.5425
},
"103": {
"Precision": 0.6082,
"Recall": 0.6898,
"Specificity": 0.9725,
"F1 score": 0.6464
},
"11": {
"Precision": 0.9167,
"Recall": 0.8462,
"Specificity": 0.9997,
"F1 score": 0.88
},
"12": {
"Precision": 1.0,
"Recall": 0.8148,
"Specificity": 1.0,
"F1 score": 0.898
},
"13": {
"Precision": 0.8077,
"Recall": 0.8182,
"Specificity": 0.9959,
"F1 score": 0.8129
},
"14": {
"Precision": 0.871,
"Recall": 0.8182,
"Specificity": 0.9978,
"F1 score": 0.8438
},
"15": {
"Precision": 0.5714,
"Recall": 0.6667,
"Specificity": 0.9992,
"F1 score": 0.6154
},
"16": {
"Precision": 0.9333,
"Recall": 0.875,
"Specificity": 0.9997,
"F1 score": 0.9032
},
"17": {
"Precision": 0.9,
"Recall": 0.6429,
"Specificity": 0.9997,
"F1 score": 0.75
},
"18": {
"Precision": 0.6071,
"Recall": 0.6538,
"Specificity": 0.997,
"F1 score": 0.6296
},
"19": {
"Precision": 0.625,
"Recall": 0.625,
"Specificity": 0.9992,
"F1 score": 0.625
},
"2": {
"Precision": 0.6667,
"Recall": 0.3333,
"Specificity": 0.9997,
"F1 score": 0.4444
},
"20": {
"Precision": 0.5,
"Recall": 0.5,
"Specificity": 0.9992,
"F1 score": 0.5
},
"21": {
"Precision": 0.6786,
"Recall": 0.6786,
"Specificity": 0.9976,
"F1 score": 0.6786
},
"22": {
"Precision": 0.8571,
"Recall": 0.8571,
"Specificity": 0.9995,
"F1 score": 0.8571
},
"23": {
"Precision": 1.0,
"Recall": 0.6667,
"Specificity": 1.0,
"F1 score": 0.8
},
"24": {
"Precision": 0.9,
"Recall": 0.75,
"Specificity": 0.9995,
"F1 score": 0.8182
},
"25": {
"Precision": 0.6667,
"Recall": 0.75,
"Specificity": 0.9976,
"F1 score": 0.7059
},
"26": {
"Precision": 0.8333,
"Recall": 0.8333,
"Specificity": 0.9997,
"F1 score": 0.8333
},
"27": {
"Precision": 1.0,
"Recall": 0.9,
"Specificity": 1.0,
"F1 score": 0.9474
},
"28": {
"Precision": 0.9615,
"Recall": 0.7353,
"Specificity": 0.9997,
"F1 score": 0.8333
},
"29": {
"Precision": 0.5833,
"Recall": 0.6562,
"Specificity": 0.9959,
"F1 score": 0.6176
},
"3": {
"Precision": 0.3333,
"Recall": 0.1667,
"Specificity": 0.9995,
"F1 score": 0.2222
},
"30": {
"Precision": 0.5357,
"Recall": 0.4839,
"Specificity": 0.9965,
"F1 score": 0.5085
},
"31": {
"Precision": 0.5714,
"Recall": 0.5714,
"Specificity": 0.9992,
"F1 score": 0.5714
},
"32": {
"Precision": 1.0,
"Recall": 0.7143,
"Specificity": 1.0,
"F1 score": 0.8333
},
"33": {
"Precision": 0.5714,
"Recall": 0.6667,
"Specificity": 0.9992,
"F1 score": 0.6154
},
"34": {
"Precision": 0.7143,
"Recall": 1.0,
"Specificity": 0.9995,
"F1 score": 0.8333
},
"35": {
"Precision": 1.0,
"Recall": 0.7273,
"Specificity": 1.0,
"F1 score": 0.8421
},
"36": {
"Precision": 0.8947,
"Recall": 1.0,
"Specificity": 0.9995,
"F1 score": 0.9444
},
"37": {
"Precision": 0.875,
"Recall": 0.875,
"Specificity": 0.9997,
"F1 score": 0.875
},
"38": {
"Precision": 1.0,
"Recall": 0.4,
"Specificity": 1.0,
"F1 score": 0.5714
},
"39": {
"Precision": 0.8571,
"Recall": 0.5714,
"Specificity": 0.9995,
"F1 score": 0.6857
},
"4": {
"Precision": 0.6858,
"Recall": 0.7561,
"Specificity": 0.9798,
"F1 score": 0.7192
},
"40": {
"Precision": 0.7059,
"Recall": 0.6667,
"Specificity": 0.9986,
"F1 score": 0.6857
},
"41": {
"Precision": 0.6774,
"Recall": 0.75,
"Specificity": 0.9973,
"F1 score": 0.7119
},
"42": {
"Precision": 1.0,
"Recall": 0.4444,
"Specificity": 1.0,
"F1 score": 0.6153
},
"43": {
"Precision": 0.8636,
"Recall": 0.5937,
"Specificity": 0.9992,
"F1 score": 0.7037
},
"44": {
"Precision": 0.8333,
"Recall": 1.0,
"Specificity": 0.9997,
"F1 score": 0.9091
},
"45": {
"Precision": 0.8085,
"Recall": 0.76,
"Specificity": 0.9975,
"F1 score": 0.7835
},
"46": {
"Precision": 0.7778,
"Recall": 0.7568,
"Specificity": 0.9978,
"F1 score": 0.7672
},
"47": {
"Precision": 0.7308,
"Recall": 0.75,
"Specificity": 0.9942,
"F1 score": 0.7403
},
"48": {
"Precision": 0.7627,
"Recall": 0.7317,
"Specificity": 0.9922,
"F1 score": 0.7469
},
"49": {
"Precision": 0.6789,
"Recall": 0.7866,
"Specificity": 0.9828,
"F1 score": 0.7288
},
"5": {
"Precision": 0.9,
"Recall": 0.72,
"Specificity": 0.9995,
"F1 score": 0.8
},
"50": {
"Precision": 0.5846,
"Recall": 0.6552,
"Specificity": 0.9926,
"F1 score": 0.6179
},
"51": {
"Precision": 0.75,
"Recall": 0.6,
"Specificity": 0.9984,
"F1 score": 0.6667
},
"52": {
"Precision": 0.5385,
"Recall": 0.6364,
"Specificity": 0.9951,
"F1 score": 0.5834
},
"53": {
"Precision": 0.8649,
"Recall": 0.7164,
"Specificity": 0.9958,
"F1 score": 0.7837
},
"54": {
"Precision": 0.8462,
"Recall": 1.0,
"Specificity": 0.9995,
"F1 score": 0.9167
},
"55": {
"Precision": 0.8421,
"Recall": 0.9412,
"Specificity": 0.9992,
"F1 score": 0.8889
},
"56": {
"Precision": 0.7727,
"Recall": 0.6538,
"Specificity": 0.9986,
"F1 score": 0.7083
},
"57": {
"Precision": 0.9444,
"Recall": 0.9444,
"Specificity": 0.9997,
"F1 score": 0.9444
},
"58": {
"Precision": 1.0,
"Recall": 1.0,
"Specificity": 1.0,
"F1 score": 1.0
},
"59": {
"Precision": 1.0,
"Recall": 1.0,
"Specificity": 1.0,
"F1 score": 1.0
},
"6": {
"Precision": 0.8,
"Recall": 0.8,
"Specificity": 0.9997,
"F1 score": 0.8
},
"60": {
"Precision": 1.0,
"Recall": 1.0,
"Specificity": 1.0,
"F1 score": 1.0
},
"61": {
"Precision": 0.5714,
"Recall": 0.5,
"Specificity": 0.9992,
"F1 score": 0.5333
},
"62": {
"Precision": 0.76,
"Recall": 0.7037,
"Specificity": 0.9984,
"F1 score": 0.7308
},
"63": {
"Precision": 0.8889,
"Recall": 1.0,
"Specificity": 0.9997,
"F1 score": 0.9412
},
"64": {
"Precision": 0.9375,
"Recall": 0.9375,
"Specificity": 0.9997,
"F1 score": 0.9375
},
"65": {
"Precision": 1.0,
"Recall": 1.0,
"Specificity": 1.0,
"F1 score": 1.0
},
"66": {
"Precision": 0.6667,
"Recall": 0.3333,
"Specificity": 0.9997,
"F1 score": 0.4444
},
"67": {
"Precision": 0.7661,
"Recall": 0.7325,
"Specificity": 0.9854,
"F1 score": 0.7489
},
"68": {
"Precision": 0.5421,
"Recall": 0.7632,
"Specificity": 0.9865,
"F1 score": 0.6339
},
"69": {
"Precision": 0.9048,
"Recall": 0.7037,
"Specificity": 0.9995,
"F1 score": 0.7917
},
"7": {
"Precision": 0.8667,
"Recall": 0.8387,
"Specificity": 0.9989,
"F1 score": 0.8525
},
"70": {
"Precision": 0.7407,
"Recall": 0.6667,
"Specificity": 0.9981,
"F1 score": 0.7018
},
"71": {
"Precision": 0.6316,
"Recall": 0.6,
"Specificity": 0.9962,
"F1 score": 0.6154
},
"72": {
"Precision": 0.7321,
"Recall": 0.8367,
"Specificity": 0.9959,
"F1 score": 0.7809
},
"73": {
"Precision": 0.5443,
"Recall": 0.6418,
"Specificity": 0.9799,
"F1 score": 0.589
},
"74": {
"Precision": 0.7105,
"Recall": 0.75,
"Specificity": 0.997,
"F1 score": 0.7297
},
"75": {
"Precision": 0.7412,
"Recall": 0.7079,
"Specificity": 0.9939,
"F1 score": 0.7242
},
"76": {
"Precision": 1.0,
"Recall": 0.8857,
"Specificity": 1.0,
"F1 score": 0.9394
},
"77": {
"Precision": 0.7568,
"Recall": 0.6829,
"Specificity": 0.9975,
"F1 score": 0.718
},
"78": {
"Precision": 0.9048,
"Recall": 0.76,
"Specificity": 0.9995,
"F1 score": 0.8261
},
"79": {
"Precision": 0.8889,
"Recall": 0.7059,
"Specificity": 0.9992,
"F1 score": 0.7869
},
"8": {
"Precision": 0.6897,
"Recall": 0.8,
"Specificity": 0.9976,
"F1 score": 0.7408
},
"80": {
"Precision": 0.9444,
"Recall": 0.7556,
"Specificity": 0.9995,
"F1 score": 0.8395
},
"81": {
"Precision": 0.8235,
"Recall": 0.4828,
"Specificity": 0.9992,
"F1 score": 0.6087
},
"82": {
"Precision": 0.6585,
"Recall": 0.6923,
"Specificity": 0.9962,
"F1 score": 0.675
},
"83": {
"Precision": 0.6786,
"Recall": 0.5758,
"Specificity": 0.9976,
"F1 score": 0.623
},
"84": {
"Precision": 1.0,
"Recall": 0.6667,
"Specificity": 1.0,
"F1 score": 0.8
},
"85": {
"Precision": 1.0,
"Recall": 0.8889,
"Specificity": 1.0,
"F1 score": 0.9412
},
"86": {
"Precision": 0.7407,
"Recall": 0.5714,
"Specificity": 0.9981,
"F1 score": 0.6451
},
"87": {
"Precision": 0.6957,
"Recall": 0.7442,
"Specificity": 0.9962,
"F1 score": 0.7191
},
"88": {
"Precision": 0.9259,
"Recall": 0.8929,
"Specificity": 0.9995,
"F1 score": 0.9091
},
"89": {
"Precision": 0.8182,
"Recall": 0.6923,
"Specificity": 0.9995,
"F1 score": 0.75
},
"9": {
"Precision": 0.84,
"Recall": 0.875,
"Specificity": 0.9989,
"F1 score": 0.8571
},
"90": {
"Precision": 0.875,
"Recall": 0.4516,
"Specificity": 0.9995,
"F1 score": 0.5957
},
"91": {
"Precision": 0.7576,
"Recall": 0.7812,
"Specificity": 0.9978,
"F1 score": 0.7692
},
"92": {
"Precision": 0.8571,
"Recall": 0.8571,
"Specificity": 0.9997,
"F1 score": 0.8571
},
"93": {
"Precision": 0.7255,
"Recall": 0.9024,
"Specificity": 0.9962,
"F1 score": 0.8043
},
"94": {
"Precision": 0.6512,
"Recall": 0.7368,
"Specificity": 0.9959,
"F1 score": 0.6914
},
"95": {
"Precision": 0.6286,
"Recall": 0.5946,
"Specificity": 0.9965,
"F1 score": 0.6111
},
"96": {
"Precision": 0.5714,
"Recall": 0.5517,
"Specificity": 0.9967,
"F1 score": 0.5614
},
"97": {
"Precision": 0.7778,
"Recall": 0.5833,
"Specificity": 0.9995,
"F1 score": 0.6667
},
"98": {
"Precision": 0.8333,
"Recall": 1.0,
"Specificity": 0.9995,
"F1 score": 0.9091
},
"99": {
"Precision": 0.7778,
"Recall": 1.0,
"Specificity": 0.9995,
"F1 score": 0.875
},
"mean precision": 0.7854182692307694,
"mean recall": 0.7331990384615386,
"mean specificity": 0.9972884615384617,
"mean f1 score": 0.7497365384615383
}
}训练集和测试集的混淆矩阵：这里类别太多了，显示的有点密集
2.4 推理
推理是指没有标签，只有图片数据的情况下对数据的预测，这里直接运行predict脚本即可
需要把待推理的数据放在 inference/img 下
3. 下载
关于本项目代码和数据集、训练结果的下载：图像分类实战：EfficientNet轻量级网络实现的迁移学习、图像识别项目：大型104种常见花种类识别资源-CSDN文库
关于图像分类网络的改进可以参考：